diff --git a/README.md b/README.md
index a7e830b11b8f67b7866abc46777b5c588245239d..7babf2c3c1236c32dbe87437adab5acb1dfb7661 100644
--- a/README.md
+++ b/README.md
@@ -1,12 +1,14 @@
 ---
-title: FlipSketch
-emoji: 🚀
-colorFrom: purple
-colorTo: green
-sdk: docker
-pinned: false
-license: mit
-short_description: Sketch Animations
+title: FlipSketch       
+emoji: 🚀                 
+colorFrom: blue            
+colorTo: green             
+sdk: gradio                
+app_file: app.py           
+pinned: false              
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+
+# FlipSketch
+
+FlipSketch: Flipping assets Drawings to Text-Guided Sketch Animations
\ No newline at end of file
diff --git a/app.py b/app.py
new file mode 100644
index 0000000000000000000000000000000000000000..744792e7d503d169732b02540bc7ac677909a4fa
--- /dev/null
+++ b/app.py
@@ -0,0 +1,271 @@
+from flask import Flask, render_template, request, jsonify
+import os
+import cv2
+import subprocess
+import torch
+import torchvision
+import warnings
+import numpy as np
+from PIL import Image, ImageSequence
+from moviepy.editor import VideoFileClip
+import imageio
+import uuid
+
+from diffusers import (
+    TextToVideoSDPipeline,
+    AutoencoderKL,
+    DDPMScheduler,
+    DDIMScheduler,
+    UNet3DConditionModel,
+)
+import time
+from transformers import CLIPTokenizer, CLIPTextModel
+
+from diffusers.utils import export_to_video
+from gifs_filter import filter
+from invert_utils import ddim_inversion as dd_inversion
+from text2vid_modded import TextToVideoSDPipelineModded
+
+
+def run_setup():
+    try:
+        # Step 1: Install Git LFS
+        subprocess.run(["git", "lfs", "install"], check=True)
+
+        # Step 2: Clone the repository
+        repo_url = "https://huggingface.co/Hmrishav/t2v_sketch-lora"
+        subprocess.run(["git", "clone", repo_url], check=True)
+
+        # Step 3: Move the checkpoint file
+        source = "t2v_sketch-lora/checkpoint-2500"
+        destination = "./checkpoint-2500/"
+        os.rename(source, destination)
+
+        print("Setup completed successfully!")
+    except subprocess.CalledProcessError as e:
+        print(f"Error during setup: {e}")
+    except FileNotFoundError as e:
+        print(f"File operation error: {e}")
+    except Exception as e:
+        print(f"Unexpected error: {e}")
+
+# Automatically run setup during app initialization
+run_setup()
+
+
+# Flask app setup
+app = Flask(__name__)
+app.config['UPLOAD_FOLDER'] = 'static/uploads'
+app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024  # 16MB max file size
+os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
+
+# Environment setup
+os.environ["TORCH_CUDNN_V8_API_ENABLED"] = "1"
+LORA_CHECKPOINT = "checkpoint-2500"
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+dtype = torch.bfloat16
+
+# Helper functions
+
+def cleanup_old_files(directory, age_in_seconds = 600):
+    """
+    Deletes files older than a certain age in the specified directory.
+
+    Args:
+        directory (str): The directory to clean up.
+        age_in_seconds (int): The age in seconds; files older than this will be deleted.
+    """
+    now = time.time()
+    for filename in os.listdir(directory):
+        file_path = os.path.join(directory, filename)
+        # Only delete files (not directories)
+        if os.path.isfile(file_path):
+            file_age = now - os.path.getmtime(file_path)
+            if file_age > age_in_seconds:
+                try:
+                    os.remove(file_path)
+                    print(f"Deleted old file: {file_path}")
+                except Exception as e:
+                    print(f"Error deleting file {file_path}: {e}")
+                    
+def load_frames(image: Image, mode='RGBA'):
+    return np.array([np.array(frame.convert(mode)) for frame in ImageSequence.Iterator(image)])
+
+def save_gif(frames, path):
+    imageio.mimsave(path, [frame.astype(np.uint8) for frame in frames], format='GIF', duration=1/10)
+
+def load_image(imgname, target_size=None):
+    pil_img = Image.open(imgname).convert('RGB')
+    if target_size:
+        if isinstance(target_size, int):
+            target_size = (target_size, target_size)
+        pil_img = pil_img.resize(target_size, Image.Resampling.LANCZOS)
+    return torchvision.transforms.ToTensor()(pil_img).unsqueeze(0)  # Add batch dimension
+
+def prepare_latents(pipe, x_aug):
+    with torch.cuda.amp.autocast():
+        batch_size, num_frames, channels, height, width = x_aug.shape
+        x_aug = x_aug.reshape(batch_size * num_frames, channels, height, width)
+        latents = pipe.vae.encode(x_aug).latent_dist.sample()
+        latents = latents.view(batch_size, num_frames, -1, latents.shape[2], latents.shape[3])
+        latents = latents.permute(0, 2, 1, 3, 4)
+    return pipe.vae.config.scaling_factor * latents
+
+@torch.no_grad()
+def invert(pipe, inv, load_name, device="cuda", dtype=torch.bfloat16):
+    input_img = [load_image(load_name, 256).to(device, dtype=dtype).unsqueeze(1)] * 5
+    input_img = torch.cat(input_img, dim=1)
+    latents = prepare_latents(pipe, input_img).to(torch.bfloat16)
+    inv.set_timesteps(25)
+    id_latents = dd_inversion(pipe, inv, video_latent=latents, num_inv_steps=25, prompt="")[-1].to(dtype)
+    return torch.mean(id_latents, dim=2, keepdim=True)
+
+def load_primary_models(pretrained_model_path):
+    return (
+        DDPMScheduler.from_config(pretrained_model_path, subfolder="scheduler"),
+        CLIPTokenizer.from_pretrained(pretrained_model_path, subfolder="tokenizer"),
+        CLIPTextModel.from_pretrained(pretrained_model_path, subfolder="text_encoder"),
+        AutoencoderKL.from_pretrained(pretrained_model_path, subfolder="vae"),
+        UNet3DConditionModel.from_pretrained(pretrained_model_path, subfolder="unet"),
+    )
+
+
+def initialize_pipeline(model: str, device: str = "cuda"):
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        scheduler, tokenizer, text_encoder, vae, unet = load_primary_models(model)
+    pipe = TextToVideoSDPipeline.from_pretrained(
+        pretrained_model_name_or_path="damo-vilab/text-to-video-ms-1.7b",
+        scheduler=scheduler,
+        tokenizer=tokenizer,
+        text_encoder=text_encoder.to(device=device, dtype=torch.bfloat16),
+        vae=vae.to(device=device, dtype=torch.bfloat16),
+        unet=unet.to(device=device, dtype=torch.bfloat16),
+    )
+    pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+    return pipe, pipe.scheduler
+
+pipe_inversion, inv = initialize_pipeline(LORA_CHECKPOINT, device)
+pipe = TextToVideoSDPipelineModded.from_pretrained(
+    pretrained_model_name_or_path="damo-vilab/text-to-video-ms-1.7b",
+    scheduler=pipe_inversion.scheduler,
+    tokenizer=pipe_inversion.tokenizer,
+    text_encoder=pipe_inversion.text_encoder,
+    vae=pipe_inversion.vae,
+    unet=pipe_inversion.unet,
+).to(device)
+
+@torch.no_grad()
+def process(num_frames, num_seeds, generator, exp_dir, load_name, caption, lambda_):
+    pipe_inversion.to(device)
+    id_latents = invert(pipe_inversion, inv, load_name).to(device, dtype=dtype)
+    latents = id_latents.repeat(num_seeds, 1, 1, 1, 1)
+    generator = [torch.Generator(device="cuda").manual_seed(i) for i in range(num_seeds)]
+    video_frames = pipe(
+        prompt=caption, 
+        negative_prompt="",
+        num_frames=num_frames,
+        num_inference_steps=25, 
+        inv_latents=latents, 
+        guidance_scale=9,
+        generator=generator,
+        lambda_=lambda_,
+    ).frames
+    try: 
+        load_name = load_name.split("/")[-1]
+    except:
+        pass
+    gifs = []
+    for seed in range(num_seeds):
+        vid_name = f"{exp_dir}/mp4_logs/vid_{load_name[:-4]}-rand{seed}.mp4"
+        gif_name = f"{exp_dir}/gif_logs/vid_{load_name[:-4]}-rand{seed}.gif"
+        video_path = export_to_video(video_frames[seed], output_video_path=vid_name)
+        VideoFileClip(vid_name).write_gif(gif_name)
+        with Image.open(gif_name) as im:
+            frames = load_frames(im)
+
+        frames_collect = np.empty((0, 1024, 1024), int)
+        for frame in frames:
+            frame = cv2.resize(frame, (1024, 1024))[:, :, :3]
+            frame = cv2.cvtColor(255 - frame, cv2.COLOR_RGB2GRAY)
+
+            _, frame = cv2.threshold(255 - frame, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+
+            frames_collect = np.append(frames_collect, [frame], axis=0)
+
+        save_gif(frames_collect, gif_name)
+        gifs.append(gif_name)
+
+    return gifs
+        
+
+def generate_gifs(filepath, prompt, num_seeds=5, lambda_=0):
+    exp_dir = "static/app_tmp"
+    os.makedirs(exp_dir, exist_ok=True)
+    gifs = process(
+        num_frames=10,
+        num_seeds=num_seeds,
+        generator=None,
+        exp_dir=exp_dir,
+        load_name=filepath,
+        caption=prompt,
+        lambda_=lambda_
+    )
+    return gifs
+
+@app.route('/')
+def index():
+    return render_template('index.html')
+
+@app.route('/generate', methods=['POST'])
+def generate():
+    
+    directories_to_clean = [
+        app.config['UPLOAD_FOLDER'],
+        'static/app_tmp/mp4_logs',
+        'static/app_tmp/gif_logs',
+        'static/app_tmp/png_logs'
+    ]
+
+    # Perform cleanup
+    os.makedirs('static/app_tmp', exist_ok=True)
+    for directory in directories_to_clean:
+        os.makedirs(directory, exist_ok=True)  # Ensure the directory exists
+        cleanup_old_files(directory)
+
+    prompt = request.form.get('prompt', '')
+    num_gifs = int(request.form.get('seeds', 3))
+    lambda_value = 1 - float(request.form.get('lambda', 0.5))
+    selected_example = request.form.get('selected_example', None)
+    file = request.files.get('image')
+
+    if not file and not selected_example:
+        return jsonify({'error': 'No image file provided or example selected'}), 400
+
+    if selected_example:
+        # Use the selected example image
+        filepath = os.path.join('static', 'examples', selected_example)
+        unique_id = None  # No need for unique ID
+    else:
+        # Save the uploaded image
+        unique_id = str(uuid.uuid4())
+        filepath = os.path.join(app.config['UPLOAD_FOLDER'], f"{unique_id}_uploaded_image.png")
+        file.save(filepath)
+
+    generated_gifs = generate_gifs(filepath, prompt, num_seeds=num_gifs, lambda_=lambda_value)
+
+    unique_id = str(uuid.uuid4())
+    # Append unique id to each gif path
+    for i in range(len(generated_gifs)):
+        os.rename(generated_gifs[i], f"{generated_gifs[i].split('.')[0]}_{unique_id}.gif")
+        generated_gifs[i] = f"{generated_gifs[i].split('.')[0]}_{unique_id}.gif"
+        # Move the generated gifs to the static folder
+        
+
+    filtered_gifs = filter(generated_gifs, filepath)
+    return jsonify({'gifs': filtered_gifs, 'prompt': prompt})
+
+if __name__ == '__main__':
+    
+
+    app.run(debug=True)
\ No newline at end of file
diff --git a/app_full.py b/app_full.py
new file mode 100644
index 0000000000000000000000000000000000000000..d4226687715d97360f4d693f37f96f077557e792
--- /dev/null
+++ b/app_full.py
@@ -0,0 +1,243 @@
+from flask import Flask, render_template, request, jsonify
+import os
+import cv2
+import torch
+import torchvision
+import warnings
+import numpy as np
+from PIL import Image, ImageSequence
+from moviepy.editor import VideoFileClip
+import imageio
+import uuid
+
+from diffusers import (
+    TextToVideoSDPipeline,
+    AutoencoderKL,
+    DDPMScheduler,
+    DDIMScheduler,
+    UNet3DConditionModel,
+)
+import time
+from transformers import CLIPTokenizer, CLIPTextModel
+
+from diffusers.utils import export_to_video
+from gifs_filter import filter
+from invert_utils import ddim_inversion as dd_inversion
+from text2vid_modded_full import TextToVideoSDPipelineModded
+
+# Flask app setup
+app = Flask(__name__)
+app.config['UPLOAD_FOLDER'] = 'static/uploads'
+app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024  # 16MB max file size
+os.makedirs(app.config['UPLOAD_FOLDER'], exist_ok=True)
+
+# Environment setup
+os.environ["TORCH_CUDNN_V8_API_ENABLED"] = "1"
+LORA_CHECKPOINT = "checkpoint-2500"
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+dtype = torch.bfloat16
+
+# Helper functions
+
+def cleanup_old_files(directory, age_in_seconds = 600):
+    """
+    Deletes files older than a certain age in the specified directory.
+
+    Args:
+        directory (str): The directory to clean up.
+        age_in_seconds (int): The age in seconds; files older than this will be deleted.
+    """
+    now = time.time()
+    for filename in os.listdir(directory):
+        file_path = os.path.join(directory, filename)
+        # Only delete files (not directories)
+        if os.path.isfile(file_path):
+            file_age = now - os.path.getmtime(file_path)
+            if file_age > age_in_seconds:
+                try:
+                    os.remove(file_path)
+                    print(f"Deleted old file: {file_path}")
+                except Exception as e:
+                    print(f"Error deleting file {file_path}: {e}")
+                    
+def load_frames(image: Image, mode='RGBA'):
+    return np.array([np.array(frame.convert(mode)) for frame in ImageSequence.Iterator(image)])
+
+def save_gif(frames, path):
+    imageio.mimsave(path, [frame.astype(np.uint8) for frame in frames], format='GIF', duration=1/10)
+
+def load_image(imgname, target_size=None):
+    pil_img = Image.open(imgname).convert('RGB')
+    if target_size:
+        if isinstance(target_size, int):
+            target_size = (target_size, target_size)
+        pil_img = pil_img.resize(target_size, Image.Resampling.LANCZOS)
+    return torchvision.transforms.ToTensor()(pil_img).unsqueeze(0)  # Add batch dimension
+
+def prepare_latents(pipe, x_aug):
+    with torch.cuda.amp.autocast():
+        batch_size, num_frames, channels, height, width = x_aug.shape
+        x_aug = x_aug.reshape(batch_size * num_frames, channels, height, width)
+        latents = pipe.vae.encode(x_aug).latent_dist.sample()
+        latents = latents.view(batch_size, num_frames, -1, latents.shape[2], latents.shape[3])
+        latents = latents.permute(0, 2, 1, 3, 4)
+    return pipe.vae.config.scaling_factor * latents
+
+@torch.no_grad()
+def invert(pipe, inv, load_name, device="cuda", dtype=torch.bfloat16):
+    input_img = [load_image(load_name, 256).to(device, dtype=dtype).unsqueeze(1)] * 5
+    input_img = torch.cat(input_img, dim=1)
+    latents = prepare_latents(pipe, input_img).to(torch.bfloat16)
+    inv.set_timesteps(25)
+    id_latents = dd_inversion(pipe, inv, video_latent=latents, num_inv_steps=25, prompt="")[-1].to(dtype)
+    return torch.mean(id_latents, dim=2, keepdim=True)
+
+def load_primary_models(pretrained_model_path):
+    return (
+        DDPMScheduler.from_config(pretrained_model_path, subfolder="scheduler"),
+        CLIPTokenizer.from_pretrained(pretrained_model_path, subfolder="tokenizer"),
+        CLIPTextModel.from_pretrained(pretrained_model_path, subfolder="text_encoder"),
+        AutoencoderKL.from_pretrained(pretrained_model_path, subfolder="vae"),
+        UNet3DConditionModel.from_pretrained(pretrained_model_path, subfolder="unet"),
+    )
+
+
+def initialize_pipeline(model: str, device: str = "cuda"):
+    with warnings.catch_warnings():
+        warnings.simplefilter("ignore")
+        scheduler, tokenizer, text_encoder, vae, unet = load_primary_models(model)
+    pipe = TextToVideoSDPipeline.from_pretrained(
+        pretrained_model_name_or_path="damo-vilab/text-to-video-ms-1.7b",
+        scheduler=scheduler,
+        tokenizer=tokenizer,
+        text_encoder=text_encoder.to(device=device, dtype=torch.bfloat16),
+        vae=vae.to(device=device, dtype=torch.bfloat16),
+        unet=unet.to(device=device, dtype=torch.bfloat16),
+    )
+    pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+    return pipe, pipe.scheduler
+
+pipe_inversion, inv = initialize_pipeline(LORA_CHECKPOINT, device)
+pipe = TextToVideoSDPipelineModded.from_pretrained(
+    pretrained_model_name_or_path="damo-vilab/text-to-video-ms-1.7b",
+    scheduler=pipe_inversion.scheduler,
+    tokenizer=pipe_inversion.tokenizer,
+    text_encoder=pipe_inversion.text_encoder,
+    vae=pipe_inversion.vae,
+    unet=pipe_inversion.unet,
+).to(device)
+
+@torch.no_grad()
+def process(num_frames, num_seeds, generator, exp_dir, load_name, caption, lambda_):
+    pipe_inversion.to(device)
+    id_latents = invert(pipe_inversion, inv, load_name).to(device, dtype=dtype)
+    latents = id_latents.repeat(num_seeds, 1, 1, 1, 1)
+    generator = [torch.Generator(device="cuda").manual_seed(i) for i in range(num_seeds)]
+    video_frames = pipe(
+        prompt=caption, 
+        negative_prompt="",
+        num_frames=num_frames,
+        num_inference_steps=25, 
+        inv_latents=latents, 
+        guidance_scale=9,
+        generator=generator,
+        lambda_=lambda_,
+    ).frames
+    try: 
+        load_name = load_name.split("/")[-1]
+    except:
+        pass
+    gifs = []
+    for seed in range(num_seeds):
+        vid_name = f"{exp_dir}/mp4_logs/vid_{load_name[:-4]}-rand{seed}.mp4"
+        gif_name = f"{exp_dir}/gif_logs/vid_{load_name[:-4]}-rand{seed}.gif"
+        video_path = export_to_video(video_frames[seed], output_video_path=vid_name)
+        VideoFileClip(vid_name).write_gif(gif_name)
+        with Image.open(gif_name) as im:
+            frames = load_frames(im)
+
+        frames_collect = np.empty((0, 1024, 1024), int)
+        for frame in frames:
+            frame = cv2.resize(frame, (1024, 1024))[:, :, :3]
+            frame = cv2.cvtColor(255 - frame, cv2.COLOR_RGB2GRAY)
+
+            _, frame = cv2.threshold(255 - frame, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
+
+            frames_collect = np.append(frames_collect, [frame], axis=0)
+
+        save_gif(frames_collect, gif_name)
+        gifs.append(gif_name)
+
+    return gifs
+        
+
+def generate_gifs(filepath, prompt, num_seeds=5, lambda_=0):
+    exp_dir = "static/app_tmp"
+    os.makedirs(exp_dir, exist_ok=True)
+    gifs = process(
+        num_frames=10,
+        num_seeds=num_seeds,
+        generator=None,
+        exp_dir=exp_dir,
+        load_name=filepath,
+        caption=prompt,
+        lambda_=lambda_
+    )
+    return gifs
+
+@app.route('/')
+def index():
+    return render_template('index.html')
+
+@app.route('/generate', methods=['POST'])
+def generate():
+    
+    directories_to_clean = [
+        app.config['UPLOAD_FOLDER'],
+        'static/app_tmp/mp4_logs',
+        'static/app_tmp/gif_logs',
+        'static/app_tmp/png_logs'
+    ]
+
+    # Perform cleanup
+    os.makedirs('static/app_tmp', exist_ok=True)
+    for directory in directories_to_clean:
+        os.makedirs(directory, exist_ok=True)  # Ensure the directory exists
+        cleanup_old_files(directory)
+
+    prompt = request.form.get('prompt', '')
+    num_gifs = int(request.form.get('seeds', 3))
+    lambda_value = 1 - float(request.form.get('lambda', 0.5))
+    selected_example = request.form.get('selected_example', None)
+    file = request.files.get('image')
+
+    if not file and not selected_example:
+        return jsonify({'error': 'No image file provided or example selected'}), 400
+
+    if selected_example:
+        # Use the selected example image
+        filepath = os.path.join('static', 'examples', selected_example)
+        unique_id = None  # No need for unique ID
+    else:
+        # Save the uploaded image
+        unique_id = str(uuid.uuid4())
+        filepath = os.path.join(app.config['UPLOAD_FOLDER'], f"{unique_id}_uploaded_image.png")
+        file.save(filepath)
+
+    generated_gifs = generate_gifs(filepath, prompt, num_seeds=num_gifs, lambda_=lambda_value)
+
+    unique_id = str(uuid.uuid4())
+    # Append unique id to each gif path
+    for i in range(len(generated_gifs)):
+        os.rename(generated_gifs[i], f"{generated_gifs[i].split('.')[0]}_{unique_id}.gif")
+        generated_gifs[i] = f"{generated_gifs[i].split('.')[0]}_{unique_id}.gif"
+        # Move the generated gifs to the static folder
+        
+
+    filtered_gifs = filter(generated_gifs, filepath)
+    return jsonify({'gifs': filtered_gifs, 'prompt': prompt})
+
+if __name__ == '__main__':
+    
+
+    app.run(debug=True)
\ No newline at end of file
diff --git a/environment.yml b/environment.yml
new file mode 100644
index 0000000000000000000000000000000000000000..1bc67341b2e9a0f485ccbb57b0932591267aa2c9
--- /dev/null
+++ b/environment.yml
@@ -0,0 +1,402 @@
+name: flipsketch
+channels:
+  - pytorch
+  - nvidia
+  - conda-forge
+  - defaults
+dependencies:
+  - _libgcc_mutex=0.1=main
+  - _openmp_mutex=5.1=1_gnu
+  - asttokens=2.4.1=pyhd8ed1ab_0
+  - blas=1.0=mkl
+  - brotli-python=1.0.9=py310hd8f1fbe_7
+  - bzip2=1.0.8=h7f98852_4
+  - ca-certificates=2024.2.2=hbcca054_0
+  - certifi=2024.2.2=pyhd8ed1ab_0
+  - charset-normalizer=2.0.4=pyhd8ed1ab_0
+  - comm=0.2.2=pyhd8ed1ab_0
+  - cuda=11.6.1=0
+  - cuda-cccl=11.6.55=hf6102b2_0
+  - cuda-command-line-tools=11.6.2=0
+  - cuda-compiler=11.6.2=0
+  - cuda-cudart=11.6.55=he381448_0
+  - cuda-cudart-dev=11.6.55=h42ad0f4_0
+  - cuda-cuobjdump=11.6.124=h2eeebcb_0
+  - cuda-cupti=11.6.124=h86345e5_0
+  - cuda-cuxxfilt=11.6.124=hecbf4f6_0
+  - cuda-driver-dev=11.6.55=0
+  - cuda-gdb=12.4.127=0
+  - cuda-libraries=11.6.1=0
+  - cuda-libraries-dev=11.6.1=0
+  - cuda-memcheck=11.8.86=0
+  - cuda-nsight=12.4.127=0
+  - cuda-nsight-compute=12.4.1=0
+  - cuda-nvcc=11.6.124=hbba6d2d_0
+  - cuda-nvdisasm=12.4.127=0
+  - cuda-nvml-dev=11.6.55=haa9ef22_0
+  - cuda-nvprof=12.4.127=0
+  - cuda-nvprune=11.6.124=he22ec0a_0
+  - cuda-nvrtc=11.6.124=h020bade_0
+  - cuda-nvrtc-dev=11.6.124=h249d397_0
+  - cuda-nvtx=11.6.124=h0630a44_0
+  - cuda-nvvp=12.4.127=0
+  - cuda-runtime=11.6.1=0
+  - cuda-samples=11.6.101=h8efea70_0
+  - cuda-sanitizer-api=12.4.127=0
+  - cuda-toolkit=11.6.1=0
+  - cuda-tools=11.6.1=0
+  - cuda-visual-tools=11.6.1=0
+  - debugpy=1.6.7=py310h6a678d5_0
+  - entrypoints=0.4=pyhd8ed1ab_0
+  - exceptiongroup=1.2.0=pyhd8ed1ab_2
+  - executing=2.0.1=pyhd8ed1ab_0
+  - ffmpeg=4.3=hf484d3e_0
+  - freetype=2.12.1=h4a9f257_0
+  - gds-tools=1.9.1.3=0
+  - gmp=6.2.1=h58526e2_0
+  - gnutls=3.6.15=he1e5248_0
+  - idna=3.4=pyhd8ed1ab_0
+  - intel-openmp=2023.1.0=hdb19cb5_46306
+  - ipykernel=6.29.3=pyhd33586a_0
+  - jedi=0.19.1=pyhd8ed1ab_0
+  - jpeg=9e=h166bdaf_1
+  - jupyter_client=7.3.4=pyhd8ed1ab_0
+  - jupyter_core=5.7.2=pyh31011fe_1
+  - lame=3.100=h7f98852_1001
+  - lcms2=2.12=h3be6417_0
+  - ld_impl_linux-64=2.38=h1181459_1
+  - lerc=3.0=h9c3ff4c_0
+  - libcublas=11.9.2.110=h5e84587_0
+  - libcublas-dev=11.9.2.110=h5c901ab_0
+  - libcufft=10.7.1.112=hf425ae0_0
+  - libcufft-dev=10.7.1.112=ha5ce4c0_0
+  - libcufile=1.9.1.3=0
+  - libcufile-dev=1.9.1.3=0
+  - libcurand=10.3.5.147=0
+  - libcurand-dev=10.3.5.147=0
+  - libcusolver=11.3.4.124=h33c3c4e_0
+  - libcusparse=11.7.2.124=h7538f96_0
+  - libcusparse-dev=11.7.2.124=hbbe9722_0
+  - libdeflate=1.17=h5eee18b_1
+  - libffi=3.4.4=h6a678d5_1
+  - libgcc-ng=11.2.0=h1234567_1
+  - libgomp=11.2.0=h1234567_1
+  - libiconv=1.16=h516909a_0
+  - libidn2=2.3.4=h5eee18b_0
+  - libnpp=11.6.3.124=hd2722f0_0
+  - libnpp-dev=11.6.3.124=h3c42840_0
+  - libnvjpeg=11.6.2.124=hd473ad6_0
+  - libnvjpeg-dev=11.6.2.124=hb5906b9_0
+  - libpng=1.6.39=h5eee18b_0
+  - libsodium=1.0.18=h36c2ea0_1
+  - libstdcxx-ng=11.2.0=he4da1e4_16
+  - libtasn1=4.19.0=h5eee18b_0
+  - libtiff=4.5.1=h6a678d5_0
+  - libunistring=0.9.10=h7f98852_0
+  - libuuid=1.41.5=h5eee18b_0
+  - libwebp-base=1.3.2=h5eee18b_0
+  - lz4-c=1.9.4=h6a678d5_1
+  - mkl=2023.1.0=h213fc3f_46344
+  - mkl-service=2.4.0=py310h5eee18b_1
+  - mkl_fft=1.3.8=py310h5eee18b_0
+  - mkl_random=1.2.4=py310hdb19cb5_0
+  - ncurses=6.4=h6a678d5_0
+  - nest-asyncio=1.6.0=pyhd8ed1ab_0
+  - nettle=3.7.3=hbbd107a_1
+  - nsight-compute=2024.1.1.4=0
+  - numpy-base=1.26.4=py310hb5e798b_0
+  - openh264=2.1.1=h780b84a_0
+  - openjpeg=2.4.0=h9ca470c_2
+  - openssl=3.0.13=h7f8727e_2
+  - packaging=24.0=pyhd8ed1ab_0
+  - parso=0.8.4=pyhd8ed1ab_0
+  - pexpect=4.9.0=pyhd8ed1ab_0
+  - pickleshare=0.7.5=py_1003
+  - pip=23.3.1=pyhd8ed1ab_0
+  - ptyprocess=0.7.0=pyhd3deb0d_0
+  - pure_eval=0.2.2=pyhd8ed1ab_0
+  - pygments=2.17.2=pyhd8ed1ab_0
+  - pysocks=1.7.1=pyha2e5f31_6
+  - python=3.10.14=h955ad1f_0
+  - python_abi=3.10=2_cp310
+  - pytorch=1.13.1=py3.10_cuda11.6_cudnn8.3.2_0
+  - pytorch-cuda=11.6=h867d48c_1
+  - pytorch-mutex=1.0=cuda
+  - pyzmq=25.1.2=py310h6a678d5_0
+  - readline=8.2=h5eee18b_0
+  - requests=2.31.0=pyhd8ed1ab_0
+  - setuptools=68.2.2=pyhd8ed1ab_0
+  - six=1.16.0=pyh6c4a22f_0
+  - sqlite=3.41.2=h5eee18b_0
+  - tbb=2021.8.0=hdb19cb5_0
+  - tk=8.6.12=h1ccaba5_0
+  - torchaudio=0.13.1=py310_cu116
+  - tornado=6.1=py310h5764c6d_3
+  - typing_extensions=4.9.0=pyha770c72_0
+  - tzdata=2024a=h8827d51_1
+  - urllib3=2.1.0=pyhd8ed1ab_0
+  - wcwidth=0.2.13=pyhd8ed1ab_0
+  - wheel=0.41.2=pyhd8ed1ab_0
+  - xz=5.4.6=h5eee18b_1
+  - zeromq=4.3.5=h6a678d5_0
+  - zlib=1.2.13=h5eee18b_1
+  - zstd=1.5.5=hc292b87_2
+  - pip:
+      - absl-py==2.1.0
+      - accelerate==0.29.2
+      - addict==2.4.0
+      - aiofiles==23.2.1
+      - aiohttp==3.9.3
+      - aiosignal==1.3.1
+      - albumentations==1.3.0
+      - aliyun-python-sdk-core==2.15.1
+      - aliyun-python-sdk-kms==2.16.2
+      - annotated-types==0.7.0
+      - antlr4-python3-runtime==4.8
+      - anyio==4.6.2.post1
+      - appdirs==1.4.4
+      - async-timeout==4.0.3
+      - attrs==23.2.0
+      - basicsr==1.4.2
+      - beautifulsoup4==4.12.3
+      - bitsandbytes==0.35.4
+      - black==21.4b2
+      - blinker==1.8.2
+      - blis==0.7.11
+      - boto3==1.34.97
+      - botocore==1.34.97
+      - bresenham==0.2.1
+      - cachetools==5.3.3
+      - captum==0.7.0
+      - catalogue==2.0.10
+      - cffi==1.16.0
+      - chardet==5.2.0
+      - click==8.1.7
+      - clip==0.1.0
+      - cloudpickle==3.0.0
+      - cmake==3.25.2
+      - compel==2.0.3
+      - confection==0.1.4
+      - contourpy==1.2.1
+      - controlnet-aux==0.0.6
+      - crcmod==1.7
+      - cryptography==42.0.7
+      - cssselect2==0.7.0
+      - cycler==0.12.1
+      - cymem==2.0.8
+      - cython==3.0.10
+      - datasets==2.18.0
+      - decorator==4.4.2
+      - decord==0.6.0
+      - deepspeed==0.8.0
+      - diffdist==0.1
+      - diffusers==0.27.2
+      - dill==0.3.8
+      - docker-pycreds==0.4.0
+      - easydict==1.10
+      - einops==0.3.0
+      - fairscale==0.4.13
+      - faiss-cpu==1.8.0
+      - fastapi==0.115.4
+      - ffmpy==0.3.0
+      - filelock==3.13.4
+      - flask==3.0.3
+      - flatbuffers==24.3.25
+      - fonttools==4.51.0
+      - frozenlist==1.4.1
+      - fsspec==2024.2.0
+      - ftfy==6.1.1
+      - future==1.0.0
+      - fvcore==0.1.5.post20221221
+      - gast==0.5.4
+      - gdown==5.1.0
+      - gitdb==4.0.11
+      - gitpython==3.1.43
+      - google-auth==2.29.0
+      - google-auth-oauthlib==0.4.6
+      - gradio==5.5.0
+      - gradio-client==1.4.2
+      - grpcio==1.62.1
+      - h11==0.14.0
+      - hjson==3.1.0
+      - httpcore==1.0.6
+      - httpx==0.27.2
+      - huggingface-hub==0.25.2
+      - hydra-core==1.1.1
+      - imageio==2.25.1
+      - imageio-ffmpeg==0.4.8
+      - importlib-metadata==7.1.0
+      - inquirerpy==0.3.4
+      - iopath==0.1.9
+      - ipdb==0.13.13
+      - ipympl==0.9.4
+      - ipython==8.23.0
+      - ipython-genutils==0.2.0
+      - ipywidgets==8.1.2
+      - itsdangerous==2.2.0
+      - jax==0.4.26
+      - jaxlib==0.4.26
+      - jinja2==3.1.3
+      - jmespath==0.10.0
+      - joblib==1.4.2
+      - jupyterlab-widgets==3.0.10
+      - kiwisolver==1.4.5
+      - kornia==0.6.0
+      - lightning-utilities==0.11.2
+      - lmdb==1.4.1
+      - loguru==0.7.2
+      - loralib==0.1.2
+      - lvis==0.5.3
+      - lxml==5.2.1
+      - markdown==3.6
+      - markdown-it-py==3.0.0
+      - markupsafe==2.1.5
+      - matplotlib==3.8.4
+      - matplotlib-inline==0.1.6
+      - mdurl==0.1.2
+      - mediapipe==0.10.11
+      - ml-dtypes==0.4.0
+      - modelcards==0.1.6
+      - modelscope==1.14.0
+      - motion-vector-extractor==1.0.6
+      - moviepy==1.0.3
+      - mpmath==1.3.0
+      - multidict==6.0.5
+      - multiprocess==0.70.16
+      - murmurhash==1.0.10
+      - mypy-extensions==1.0.0
+      - networkx==3.3
+      - ninja==1.11.1.1
+      - nltk==3.8.1
+      - numpy==1.24.2
+      - nvidia-cublas-cu11==11.10.3.66
+      - nvidia-cuda-cupti-cu12==12.1.105
+      - nvidia-cuda-nvrtc-cu11==11.7.99
+      - nvidia-cuda-nvrtc-cu12==12.1.105
+      - nvidia-cuda-runtime-cu11==11.7.99
+      - nvidia-cuda-runtime-cu12==12.1.105
+      - nvidia-cudnn-cu11==8.5.0.96
+      - nvidia-cufft-cu12==11.0.2.54
+      - nvidia-curand-cu12==10.3.2.106
+      - nvidia-nccl-cu12==2.20.5
+      - nvidia-nvjitlink-cu12==12.6.77
+      - nvidia-nvtx-cu12==12.1.105
+      - oauthlib==3.2.2
+      - omegaconf==2.1.1
+      - open-clip-torch==2.0.2
+      - opencv-contrib-python==4.9.0.80
+      - opencv-python==4.6.0.66
+      - opencv-python-headless==4.9.0.80
+      - opt-einsum==3.3.0
+      - orjson==3.10.11
+      - oss2==2.18.5
+      - pandas==1.5.3
+      - pathspec==0.12.1
+      - pathtools==0.1.2
+      - peft==0.10.0
+      - pfzy==0.3.4
+      - pillow==9.5.0
+      - pkgconfig==1.5.5
+      - platformdirs==4.2.0
+      - portalocker==2.8.2
+      - preshed==3.0.9
+      - proglog==0.1.10
+      - prompt-toolkit==3.0.43
+      - protobuf==3.20.3
+      - psutil==5.9.8
+      - py-cpuinfo==9.0.0
+      - pyarrow==15.0.2
+      - pyarrow-hotfix==0.6
+      - pyasn1==0.6.0
+      - pyasn1-modules==0.4.0
+      - pyav==12.0.5
+      - pycocotools==2.0.7
+      - pycparser==2.22
+      - pycryptodome==3.20.0
+      - pydantic==2.9.2
+      - pydantic-core==2.23.4
+      - pydeprecate==0.3.1
+      - pydot==2.0.0
+      - pydub==0.25.1
+      - pynvml==11.5.3
+      - pyparsing==3.1.2
+      - pyre-extensions==0.0.23
+      - python-dateutil==2.9.0.post0
+      - python-multipart==0.0.12
+      - pytorch-lightning==1.4.2
+      - pytz==2024.1
+      - pywavelets==1.6.0
+      - pyyaml==6.0.1
+      - qudida==0.0.4
+      - regex==2024.4.16
+      - reportlab==4.1.0
+      - requests-oauthlib==2.0.0
+      - rich==13.9.4
+      - rsa==4.9
+      - ruff==0.7.2
+      - s3transfer==0.10.1
+      - safehttpx==0.1.1
+      - safetensors==0.4.2
+      - scikit-image==0.19.3
+      - scikit-learn==1.4.2
+      - scikit-video==1.1.11
+      - scipy==1.10.1
+      - semantic-version==2.10.0
+      - sentry-sdk==1.44.1
+      - setproctitle==1.3.3
+      - shapely==2.0.3
+      - shellingham==1.5.4
+      - simplejson==3.19.2
+      - smmap==5.0.1
+      - sniffio==1.3.1
+      - sortedcontainers==2.4.0
+      - sounddevice==0.4.6
+      - soupsieve==2.5
+      - srsly==2.4.8
+      - stable-diffusion-sdkit==2.1.3
+      - stack-data==0.6.3
+      - starlette==0.41.2
+      - svg-path==6.3
+      - svglib==1.5.1
+      - svgpathtools==1.6.1
+      - svgwrite==1.4.3
+      - sympy==1.13.3
+      - tabulate==0.9.0
+      - tb-nightly==2.17.0a20240408
+      - tensorboard==2.12.0
+      - tensorboard-data-server==0.7.0
+      - tensorboard-plugin-wit==1.8.1
+      - termcolor==2.2.0
+      - test-tube==0.7.5
+      - thinc==8.1.10
+      - threadpoolctl==3.5.0
+      - tifffile==2024.2.12
+      - timm==0.6.11
+      - tinycss2==1.2.1
+      - tokenizers==0.20.1
+      - toml==0.10.2
+      - tomli==2.0.1
+      - tomlkit==0.12.0
+      - torch==1.13.1
+      - torchmetrics==0.6.0
+      - torchsummary==1.5.1
+      - torchvision==0.14.1
+      - tqdm==4.64.1
+      - traitlets==5.14.2
+      - transformers==4.45.2
+      - triton==2.3.0
+      - typer==0.12.5
+      - typing-inspect==0.9.0
+      - uvicorn==0.32.0
+      - wandb==0.16.6
+      - wasabi==1.1.2
+      - webencodings==0.5.1
+      - websockets==12.0
+      - werkzeug==3.0.2
+      - widgetsnbextension==4.0.10
+      - xformers==0.0.16
+      - xxhash==3.4.1
+      - yacs==0.1.8
+      - yapf==0.40.2
+      - yarl==1.9.4
+      - zipp==3.18.1
diff --git a/gifs_filter.py b/gifs_filter.py
new file mode 100644
index 0000000000000000000000000000000000000000..ecaa366d0b1c163e893682300081d8a6d4960ff1
--- /dev/null
+++ b/gifs_filter.py
@@ -0,0 +1,68 @@
+# filter images
+from PIL import Image, ImageSequence
+import requests
+from tqdm import tqdm
+import numpy as np
+import torch
+from transformers import CLIPProcessor, CLIPModel
+
+def sample_frame_indices(clip_len, frame_sample_rate, seg_len):
+    converted_len = int(clip_len * frame_sample_rate)
+    end_idx = np.random.randint(converted_len, seg_len)
+    start_idx = end_idx - converted_len
+    indices = np.linspace(start_idx, end_idx, num=clip_len)
+    indices = np.clip(indices, start_idx, end_idx - 1).astype(np.int64)
+    return indices
+
+def load_frames(image: Image, mode='RGBA'):
+    return np.array([
+        np.array(frame.convert(mode))
+        for frame in ImageSequence.Iterator(image)
+    ])
+
+img_model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
+img_processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
+
+
+
+def filter(gifs, input_image): 
+    max_cosine = 0.9
+    max_gif = []
+    
+    for gif in tqdm(gifs, total=len(gifs)):
+        with Image.open(gif) as im:
+            frames = load_frames(im)
+
+        frames = np.array(frames)
+        frames = frames[:, :, :, :3]
+        frames = np.transpose(frames, (0, 3, 1, 2))[1:]
+
+
+
+        image = Image.open(input_image)
+        
+        
+        inputs = img_processor(images=frames, return_tensors="pt", padding=False)
+        inputs_base = img_processor(images=image, return_tensors="pt", padding=False)
+        
+        with torch.no_grad():
+            feat_img_base = img_model.get_image_features(pixel_values=inputs_base["pixel_values"])        
+            feat_img_vid = img_model.get_image_features(pixel_values=inputs["pixel_values"])
+        cos_avg = 0
+        avg_score_for_vid = 0
+        for i in range(len(feat_img_vid)):
+                
+            cosine_similarity = torch.nn.functional.cosine_similarity(
+                feat_img_base, 
+                feat_img_vid[0].unsqueeze(0), 
+                dim=1)
+            # print(cosine_similarity)
+            cos_avg += cosine_similarity.item()
+            
+        cos_avg /= len(feat_img_vid)
+        print("Current cosine similarity: ", cos_avg)
+        print("Max cosine similarity: ", max_cosine)
+        if cos_avg > max_cosine:
+            # max_cosine = cos_avg
+            max_gif.append(gif)
+    return max_gif
\ No newline at end of file
diff --git a/invert_utils.py b/invert_utils.py
new file mode 100644
index 0000000000000000000000000000000000000000..802fb8d001c7cc6f24afd07ceff7e2db3f039e5b
--- /dev/null
+++ b/invert_utils.py
@@ -0,0 +1,89 @@
+import os
+import imageio
+import numpy as np
+from typing import Union
+
+import torch
+import torchvision
+
+from tqdm import tqdm
+from einops import rearrange
+
+
+def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=4, fps=8):
+    videos = rearrange(videos, "b c t h w -> t b c h w")
+    outputs = []
+    for x in videos:
+        x = torchvision.utils.make_grid(x, nrow=n_rows)
+        x = x.transpose(0, 1).transpose(1, 2).squeeze(-1)
+        if rescale:
+            x = (x + 1.0) / 2.0  # -1,1 -> 0,1
+        x = (x * 255).numpy().astype(np.uint8)
+        outputs.append(x)
+
+    os.makedirs(os.path.dirname(path), exist_ok=True)
+    imageio.mimsave(path, outputs, fps=fps)
+
+
+# DDIM Inversion
+@torch.no_grad()
+def init_prompt(prompt, pipeline):
+    uncond_input = pipeline.tokenizer(
+        [""], padding="max_length", max_length=pipeline.tokenizer.model_max_length,
+        return_tensors="pt"
+    )
+    uncond_embeddings = pipeline.text_encoder(uncond_input.input_ids.to(pipeline.device))[0]
+    text_input = pipeline.tokenizer(
+        [prompt],
+        padding="max_length",
+        max_length=pipeline.tokenizer.model_max_length,
+        truncation=True,
+        return_tensors="pt",
+    )
+    text_embeddings = pipeline.text_encoder(text_input.input_ids.to(pipeline.device))[0]
+    context = torch.cat([uncond_embeddings, text_embeddings])
+
+    return context
+
+
+def next_step(model_output: Union[torch.FloatTensor, np.ndarray], timestep: int,
+              sample: Union[torch.FloatTensor, np.ndarray], ddim_scheduler):
+    timestep, next_timestep = min(
+        timestep - ddim_scheduler.config.num_train_timesteps // ddim_scheduler.num_inference_steps, 999), timestep
+    # try: 
+    alpha_prod_t = ddim_scheduler.alphas_cumprod[timestep] if timestep >= 0 else ddim_scheduler.final_alpha_cumprod
+    # except:
+    #     alpha_prod_t = ddim_scheduler.alphas_cumprod[timestep] #if timestep >= 0 else ddim_scheduler.final_alpha_cumprod
+    alpha_prod_t_next = ddim_scheduler.alphas_cumprod[next_timestep]
+    beta_prod_t = 1 - alpha_prod_t
+    next_original_sample = (sample - beta_prod_t ** 0.5 * model_output) / alpha_prod_t ** 0.5
+    next_sample_direction = (1 - alpha_prod_t_next) ** 0.5 * model_output
+    next_sample = alpha_prod_t_next ** 0.5 * next_original_sample + next_sample_direction
+    return next_sample
+
+
+def get_noise_pred_single(latents, t, context, unet):
+    noise_pred = unet(latents, t, encoder_hidden_states=context)["sample"]
+    return noise_pred
+
+
+@torch.no_grad()
+def ddim_loop(pipeline, ddim_scheduler, latent, num_inv_steps, prompt):
+    context = init_prompt(prompt, pipeline)
+    uncond_embeddings, cond_embeddings = context.chunk(2)
+    all_latent = [latent]
+    latent = latent.clone().detach()
+    for i in tqdm(range(num_inv_steps)):
+        t = ddim_scheduler.timesteps[len(ddim_scheduler.timesteps) - i - 1]
+        noise_pred = get_noise_pred_single(latent, t, cond_embeddings, pipeline.unet)
+        noise_pred_unc = get_noise_pred_single(latent, t, uncond_embeddings, pipeline.unet)
+        noise_pred = noise_pred_unc + 9.0 * (noise_pred_unc - noise_pred)
+        latent = next_step(noise_pred, t, latent, ddim_scheduler)
+        all_latent.append(latent)
+    return all_latent
+
+
+@torch.no_grad()
+def ddim_inversion(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt=""):
+    ddim_latents = ddim_loop(pipeline, ddim_scheduler, video_latent, num_inv_steps, prompt)
+    return ddim_latents
\ No newline at end of file
diff --git a/read_vids.py b/read_vids.py
new file mode 100644
index 0000000000000000000000000000000000000000..e9b0932e5d0a30dddd266415f9217e50b164fc23
--- /dev/null
+++ b/read_vids.py
@@ -0,0 +1,27 @@
+import imageio.v3 as iio
+import os
+from sys import argv
+video_name = argv[1]
+
+video = video_name
+video_id = video.split("/")[-1].replace(".mp4","")
+
+
+png_base = "png_logs"
+try:
+    os.mkdir(png_base)
+except:
+    pass
+
+video_id = os.path.join(png_base, video_id)
+all_frames = list(iio.imiter(video))
+
+ctr = 0
+try: 
+    os.makedirs(video_id)
+except:
+    pass
+for idx, frame in enumerate(all_frames):
+    
+    iio.imwrite(f"{video_id}/{ctr:03d}.jpg", frame)
+    ctr += 1
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000000000000000000000000000000000000..75a1f342375829525f9a979a3849fd3caf40e9f2
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,44 @@
+accelerate==0.29.2
+blinker==1.9.0
+certifi==2024.8.30
+charset-normalizer==3.4.0
+click==8.1.7
+decorator==4.4.2
+diffusers==0.27.2
+einops==0.8.0
+filelock==3.16.1
+Flask==3.0.3
+fsspec==2024.10.0
+huggingface-hub==0.25.2
+idna==3.10
+imageio==2.36.0
+imageio-ffmpeg==0.5.1
+importlib_metadata==8.5.0
+itsdangerous==2.2.0
+Jinja2==3.1.4
+MarkupSafe==3.0.2
+moviepy==1.0.3
+numpy==1.24.2
+nvidia-cublas-cu11==11.10.3.66
+nvidia-cuda-nvrtc-cu11==11.7.99
+nvidia-cuda-runtime-cu11==11.7.99
+nvidia-cudnn-cu11==8.5.0.96
+opencv-python==4.10.0.84
+packaging==24.2
+pillow==10.4.0
+proglog==0.1.10
+psutil==6.1.0
+python-dotenv==1.0.1
+PyYAML==6.0.2
+regex==2024.11.6
+requests==2.32.3
+safetensors==0.4.5
+tokenizers==0.20.3
+torch==1.13.1
+torchvision==0.14.1
+tqdm==4.67.0
+transformers==4.45.2
+typing_extensions==4.12.2
+urllib3==2.2.3
+Werkzeug==3.1.3
+zipp==3.21.0
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand0_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif b/static/app_tmp/gif_logs/vid_sketch10-rand0_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..11b5a4f4d65bd6071ff8185edf8d934f9a5c29c6
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand0_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand0_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand0_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..79b36dd8ac1ebdcd43251f6ca2e5460156965d54
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand0_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand0_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif b/static/app_tmp/gif_logs/vid_sketch10-rand0_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif
new file mode 100644
index 0000000000000000000000000000000000000000..53dd1632b4d9fa2ca068924cdd085cf47e14b241
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand0_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand0_dfcba486-0d8c-4d68-9689-97f1fb889213.gif b/static/app_tmp/gif_logs/vid_sketch10-rand0_dfcba486-0d8c-4d68-9689-97f1fb889213.gif
new file mode 100644
index 0000000000000000000000000000000000000000..6290344646e5d5920a4f5ab3e8553ec7af83df11
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand0_dfcba486-0d8c-4d68-9689-97f1fb889213.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand1_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif b/static/app_tmp/gif_logs/vid_sketch10-rand1_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..bcd6e1f1dc508a0c873c2914fadc137f18d7a2e2
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand1_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand1_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand1_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..79e6156e8a9f975606a9f8a8954ec82d830d2be9
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand1_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand1_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif b/static/app_tmp/gif_logs/vid_sketch10-rand1_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif
new file mode 100644
index 0000000000000000000000000000000000000000..d3538b7753561da3fef372dfc29c177551719679
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand1_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand1_dfcba486-0d8c-4d68-9689-97f1fb889213.gif b/static/app_tmp/gif_logs/vid_sketch10-rand1_dfcba486-0d8c-4d68-9689-97f1fb889213.gif
new file mode 100644
index 0000000000000000000000000000000000000000..495a75fc1e24756868e6306c7aa3e1aa2fcc3c61
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand1_dfcba486-0d8c-4d68-9689-97f1fb889213.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand2_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif b/static/app_tmp/gif_logs/vid_sketch10-rand2_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..90134b819d8a93ea6ac64371ff7312f83fc82943
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand2_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand2_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand2_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..d1e921c261df1568cc7473c6c69fb7736c845030
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand2_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand2_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif b/static/app_tmp/gif_logs/vid_sketch10-rand2_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif
new file mode 100644
index 0000000000000000000000000000000000000000..3f475f631334381e2f7f52818c5c492ccf06fc8a
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand2_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand2_dfcba486-0d8c-4d68-9689-97f1fb889213.gif b/static/app_tmp/gif_logs/vid_sketch10-rand2_dfcba486-0d8c-4d68-9689-97f1fb889213.gif
new file mode 100644
index 0000000000000000000000000000000000000000..a43584a102c91c401d538df0082034d3d813321d
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand2_dfcba486-0d8c-4d68-9689-97f1fb889213.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand3_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif b/static/app_tmp/gif_logs/vid_sketch10-rand3_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..064c9d832886d0259b5b8e77eb5d835c0ea007c9
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand3_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand3_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand3_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..512225ba16a81c83d5d2b2b33f85041341e522a6
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand3_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand3_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif b/static/app_tmp/gif_logs/vid_sketch10-rand3_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif
new file mode 100644
index 0000000000000000000000000000000000000000..fbdb3ef671e488b34d1e488722dcc97a4b9e8d77
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand3_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand3_dfcba486-0d8c-4d68-9689-97f1fb889213.gif b/static/app_tmp/gif_logs/vid_sketch10-rand3_dfcba486-0d8c-4d68-9689-97f1fb889213.gif
new file mode 100644
index 0000000000000000000000000000000000000000..ced90e8e998918a1f39bebffef964c10f2196596
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand3_dfcba486-0d8c-4d68-9689-97f1fb889213.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand4_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif b/static/app_tmp/gif_logs/vid_sketch10-rand4_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif
new file mode 100644
index 0000000000000000000000000000000000000000..5b4045f63f78191377749364036ba9276cb8cb0d
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand4_4e766a8e-9d22-4818-8991-e884ce17e5e5.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand4_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand4_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..124ab53ce653dd1f6062df8a80bea644ab7c7224
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand4_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand4_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif b/static/app_tmp/gif_logs/vid_sketch10-rand4_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif
new file mode 100644
index 0000000000000000000000000000000000000000..209c09fb2468c51fcbb210ab56df2071607dffe3
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand4_9e7e07af-2adc-47b0-8aa4-716a934690e8.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand4_dfcba486-0d8c-4d68-9689-97f1fb889213.gif b/static/app_tmp/gif_logs/vid_sketch10-rand4_dfcba486-0d8c-4d68-9689-97f1fb889213.gif
new file mode 100644
index 0000000000000000000000000000000000000000..df4fd44c56e174c943c7741f52352639fc6f657e
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand4_dfcba486-0d8c-4d68-9689-97f1fb889213.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand5_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand5_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..049cacdbddc16d194b073a9ce65035e4896861c3
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand5_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand6_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand6_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..8611fc3ca65fa76f7b972c33d12b4325a53d710b
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand6_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand7_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand7_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..7eda0e79cc3183ace45203a6bafb2ac4bd1c51a7
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand7_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand8_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand8_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..0d3fb01d18bd20598ce1b06cf7aea9deba07ae1d
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand8_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch10-rand9_508fa599-d685-462e-ad06-11ca4fd15d6f.gif b/static/app_tmp/gif_logs/vid_sketch10-rand9_508fa599-d685-462e-ad06-11ca4fd15d6f.gif
new file mode 100644
index 0000000000000000000000000000000000000000..46ac2c831770b6b161f03f373009308e55d4e502
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch10-rand9_508fa599-d685-462e-ad06-11ca4fd15d6f.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch3-rand0_875203a1-f830-46e7-a287-4a0bc2c3a648.gif b/static/app_tmp/gif_logs/vid_sketch3-rand0_875203a1-f830-46e7-a287-4a0bc2c3a648.gif
new file mode 100644
index 0000000000000000000000000000000000000000..ca0b791e7bf7923bd909226372022594a64cdc2b
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch3-rand0_875203a1-f830-46e7-a287-4a0bc2c3a648.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch3-rand1_875203a1-f830-46e7-a287-4a0bc2c3a648.gif b/static/app_tmp/gif_logs/vid_sketch3-rand1_875203a1-f830-46e7-a287-4a0bc2c3a648.gif
new file mode 100644
index 0000000000000000000000000000000000000000..d48e4fa3961ed8cd812c0fa743e29c7ff35c4661
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch3-rand1_875203a1-f830-46e7-a287-4a0bc2c3a648.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch3-rand2_875203a1-f830-46e7-a287-4a0bc2c3a648.gif b/static/app_tmp/gif_logs/vid_sketch3-rand2_875203a1-f830-46e7-a287-4a0bc2c3a648.gif
new file mode 100644
index 0000000000000000000000000000000000000000..e4d59f3e73c0930db699e247d78c27c4811678e6
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch3-rand2_875203a1-f830-46e7-a287-4a0bc2c3a648.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch3-rand3_875203a1-f830-46e7-a287-4a0bc2c3a648.gif b/static/app_tmp/gif_logs/vid_sketch3-rand3_875203a1-f830-46e7-a287-4a0bc2c3a648.gif
new file mode 100644
index 0000000000000000000000000000000000000000..abfb077d01dcc1e1356d41f94c7ccbafd6d7902d
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch3-rand3_875203a1-f830-46e7-a287-4a0bc2c3a648.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch3-rand4_875203a1-f830-46e7-a287-4a0bc2c3a648.gif b/static/app_tmp/gif_logs/vid_sketch3-rand4_875203a1-f830-46e7-a287-4a0bc2c3a648.gif
new file mode 100644
index 0000000000000000000000000000000000000000..be6b85bbfa89f568a99fa729307f71f29265505c
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch3-rand4_875203a1-f830-46e7-a287-4a0bc2c3a648.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand0_47fc0372-4688-4a2a-abb3-817ccfee8816.gif b/static/app_tmp/gif_logs/vid_sketch8-rand0_47fc0372-4688-4a2a-abb3-817ccfee8816.gif
new file mode 100644
index 0000000000000000000000000000000000000000..e88a745d864806273987648011af3602c40458d9
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand0_47fc0372-4688-4a2a-abb3-817ccfee8816.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand0_77158110-9239-4771-bb44-a83c3aa47567.gif b/static/app_tmp/gif_logs/vid_sketch8-rand0_77158110-9239-4771-bb44-a83c3aa47567.gif
new file mode 100644
index 0000000000000000000000000000000000000000..19ab20e3055681b3bdd5d3a404c16607d4698e2b
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand0_77158110-9239-4771-bb44-a83c3aa47567.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand0_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif b/static/app_tmp/gif_logs/vid_sketch8-rand0_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif
new file mode 100644
index 0000000000000000000000000000000000000000..630b17e0470cdfda3c9a4e31d3ca4b60fbf9b939
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand0_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand1_47fc0372-4688-4a2a-abb3-817ccfee8816.gif b/static/app_tmp/gif_logs/vid_sketch8-rand1_47fc0372-4688-4a2a-abb3-817ccfee8816.gif
new file mode 100644
index 0000000000000000000000000000000000000000..e4641d9b9ea06c027a8735df3b54ae43ef946ab2
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand1_47fc0372-4688-4a2a-abb3-817ccfee8816.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand1_77158110-9239-4771-bb44-a83c3aa47567.gif b/static/app_tmp/gif_logs/vid_sketch8-rand1_77158110-9239-4771-bb44-a83c3aa47567.gif
new file mode 100644
index 0000000000000000000000000000000000000000..f99c40b39c67111bafd799cfd196259b07441d35
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand1_77158110-9239-4771-bb44-a83c3aa47567.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand1_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif b/static/app_tmp/gif_logs/vid_sketch8-rand1_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif
new file mode 100644
index 0000000000000000000000000000000000000000..76f1c0ca11e8e90d685d27bc065cf6de68742936
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand1_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand2_47fc0372-4688-4a2a-abb3-817ccfee8816.gif b/static/app_tmp/gif_logs/vid_sketch8-rand2_47fc0372-4688-4a2a-abb3-817ccfee8816.gif
new file mode 100644
index 0000000000000000000000000000000000000000..7fa2543f6499c7626369ff4959ffbfa788c8ad38
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand2_47fc0372-4688-4a2a-abb3-817ccfee8816.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand2_77158110-9239-4771-bb44-a83c3aa47567.gif b/static/app_tmp/gif_logs/vid_sketch8-rand2_77158110-9239-4771-bb44-a83c3aa47567.gif
new file mode 100644
index 0000000000000000000000000000000000000000..a5e5a98ebb1665b9f5e387d92e47e326fcc0c137
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand2_77158110-9239-4771-bb44-a83c3aa47567.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand2_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif b/static/app_tmp/gif_logs/vid_sketch8-rand2_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif
new file mode 100644
index 0000000000000000000000000000000000000000..d45ab522052ee3362353568a019c7ac73ee18926
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand2_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand3_47fc0372-4688-4a2a-abb3-817ccfee8816.gif b/static/app_tmp/gif_logs/vid_sketch8-rand3_47fc0372-4688-4a2a-abb3-817ccfee8816.gif
new file mode 100644
index 0000000000000000000000000000000000000000..cb9d52e01ab26f74d89b548a6824f5325a95fba4
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand3_47fc0372-4688-4a2a-abb3-817ccfee8816.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand3_77158110-9239-4771-bb44-a83c3aa47567.gif b/static/app_tmp/gif_logs/vid_sketch8-rand3_77158110-9239-4771-bb44-a83c3aa47567.gif
new file mode 100644
index 0000000000000000000000000000000000000000..968cbc3d30813ee29bcdfa7e38d158a038b37278
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand3_77158110-9239-4771-bb44-a83c3aa47567.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand3_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif b/static/app_tmp/gif_logs/vid_sketch8-rand3_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif
new file mode 100644
index 0000000000000000000000000000000000000000..145814d990e5d7b477510d2bb174391af7579e97
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand3_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand4_47fc0372-4688-4a2a-abb3-817ccfee8816.gif b/static/app_tmp/gif_logs/vid_sketch8-rand4_47fc0372-4688-4a2a-abb3-817ccfee8816.gif
new file mode 100644
index 0000000000000000000000000000000000000000..758e0c617aedaf3e7e58d3a3f255348a7c535018
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand4_47fc0372-4688-4a2a-abb3-817ccfee8816.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand4_77158110-9239-4771-bb44-a83c3aa47567.gif b/static/app_tmp/gif_logs/vid_sketch8-rand4_77158110-9239-4771-bb44-a83c3aa47567.gif
new file mode 100644
index 0000000000000000000000000000000000000000..eec80d0126c125ae9b437d4e7d2c024e446421b8
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand4_77158110-9239-4771-bb44-a83c3aa47567.gif differ
diff --git a/static/app_tmp/gif_logs/vid_sketch8-rand4_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif b/static/app_tmp/gif_logs/vid_sketch8-rand4_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif
new file mode 100644
index 0000000000000000000000000000000000000000..5a2f2287a572959708be6f7850645ec58859e38f
Binary files /dev/null and b/static/app_tmp/gif_logs/vid_sketch8-rand4_fd1dace5-80a2-4a0f-afb1-c6aa0943c91a.gif differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand0.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand0.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..3702002fd69fa455ba85f4ddf8d3d003f62c1a87
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand0.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand1.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand1.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..5808fde6dde0f637b2e7acde0f6da0ae3a669099
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand1.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand2.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand2.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..18e43eae9778bb817d2fe6a088a99e7359a6b8a5
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand2.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand3.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand3.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..0c7e2b924057598fb6fb3e10e487d95b92366fc2
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand3.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand4.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand4.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..7a736b91d70771edb4cf817e97de52100fffa6a5
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand4.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand5.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand5.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..49b1c4740f693e67eaf1c6172795b6f6f1513959
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand5.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand6.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand6.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..ced1bb951c8605e49cca3d1f17a0cd294fde8399
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand6.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand7.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand7.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..21a025e950da196d6a88cf56365b29a999e73ee3
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand7.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand8.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand8.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..03177283e87822772a4a9c0898b19af18985efcb
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand8.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch10-rand9.mp4 b/static/app_tmp/mp4_logs/vid_sketch10-rand9.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..12af41f14779303e7a42b6cf6626f9c13e57abc6
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch10-rand9.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch3-rand0.mp4 b/static/app_tmp/mp4_logs/vid_sketch3-rand0.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..9d58c72dad05428636a4591131ed380726cf4675
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch3-rand0.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch3-rand1.mp4 b/static/app_tmp/mp4_logs/vid_sketch3-rand1.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..831e0d7f7eac6a0630d328a8e0165ee97fd9a068
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch3-rand1.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch3-rand2.mp4 b/static/app_tmp/mp4_logs/vid_sketch3-rand2.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..8cc33b5b713db72861cb93160288beeaa48cc213
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch3-rand2.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch3-rand3.mp4 b/static/app_tmp/mp4_logs/vid_sketch3-rand3.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..2cc9252dfe718dc671db2332adf566cf7cb1faa8
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch3-rand3.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch3-rand4.mp4 b/static/app_tmp/mp4_logs/vid_sketch3-rand4.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..1a109ffbad6f8322b8fa407d9a78dcbe8564519d
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch3-rand4.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch8-rand0.mp4 b/static/app_tmp/mp4_logs/vid_sketch8-rand0.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..e1668d3017cdd9b8681594bca5fe6f914620fcfb
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch8-rand0.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch8-rand1.mp4 b/static/app_tmp/mp4_logs/vid_sketch8-rand1.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b0e553690fd227a5fccbb359d254850e6c08f7cd
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch8-rand1.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch8-rand2.mp4 b/static/app_tmp/mp4_logs/vid_sketch8-rand2.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..29bc34c1477b6588e5f0fc6f96322cc68ed6dba5
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch8-rand2.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch8-rand3.mp4 b/static/app_tmp/mp4_logs/vid_sketch8-rand3.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..b5ad35a48a2c2963a2952cfd5ffc7feb2f234dfc
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch8-rand3.mp4 differ
diff --git a/static/app_tmp/mp4_logs/vid_sketch8-rand4.mp4 b/static/app_tmp/mp4_logs/vid_sketch8-rand4.mp4
new file mode 100644
index 0000000000000000000000000000000000000000..d619c6a310135d94799eb1dc3ee5a9bb75a942de
Binary files /dev/null and b/static/app_tmp/mp4_logs/vid_sketch8-rand4.mp4 differ
diff --git a/static/examples/sketch1.png b/static/examples/sketch1.png
new file mode 100644
index 0000000000000000000000000000000000000000..0583c7283e9336ba24389966c5290d62efc1b235
Binary files /dev/null and b/static/examples/sketch1.png differ
diff --git a/static/examples/sketch10.png b/static/examples/sketch10.png
new file mode 100644
index 0000000000000000000000000000000000000000..daa561f185de7271ba3275d2260b3b89d42321bd
Binary files /dev/null and b/static/examples/sketch10.png differ
diff --git a/static/examples/sketch11.png b/static/examples/sketch11.png
new file mode 100644
index 0000000000000000000000000000000000000000..9118593530197a56b46585edc4acf960ef0b2d9e
Binary files /dev/null and b/static/examples/sketch11.png differ
diff --git a/static/examples/sketch12.png b/static/examples/sketch12.png
new file mode 100644
index 0000000000000000000000000000000000000000..b55361e2da7d5fed2e9c942580c0429c75436b56
Binary files /dev/null and b/static/examples/sketch12.png differ
diff --git a/static/examples/sketch13.png b/static/examples/sketch13.png
new file mode 100644
index 0000000000000000000000000000000000000000..e31425a331056ee54107dec2284307f2fe5bbbe4
Binary files /dev/null and b/static/examples/sketch13.png differ
diff --git a/static/examples/sketch15.png b/static/examples/sketch15.png
new file mode 100644
index 0000000000000000000000000000000000000000..b8045c964f617f5b1adc454f8d8ee13fc38cb4e4
Binary files /dev/null and b/static/examples/sketch15.png differ
diff --git a/static/examples/sketch2.png b/static/examples/sketch2.png
new file mode 100644
index 0000000000000000000000000000000000000000..e20cf884bf84a4b789dc4e344f6ec43c689a4c3e
Binary files /dev/null and b/static/examples/sketch2.png differ
diff --git a/static/examples/sketch3.png b/static/examples/sketch3.png
new file mode 100644
index 0000000000000000000000000000000000000000..e53744ea85a3cb4ebd06279500fbe8aea85e040a
Binary files /dev/null and b/static/examples/sketch3.png differ
diff --git a/static/examples/sketch4.png b/static/examples/sketch4.png
new file mode 100644
index 0000000000000000000000000000000000000000..825e2b96e01ef527df98ef753f1247c133d64f94
Binary files /dev/null and b/static/examples/sketch4.png differ
diff --git a/static/examples/sketch5.png b/static/examples/sketch5.png
new file mode 100644
index 0000000000000000000000000000000000000000..b21e77dd93c8be87fb60e8b05aea0e9044c36ff7
Binary files /dev/null and b/static/examples/sketch5.png differ
diff --git a/static/examples/sketch6.png b/static/examples/sketch6.png
new file mode 100644
index 0000000000000000000000000000000000000000..7372989ad7c68a51821afac78cc28bbea9c95816
Binary files /dev/null and b/static/examples/sketch6.png differ
diff --git a/static/examples/sketch7.png b/static/examples/sketch7.png
new file mode 100644
index 0000000000000000000000000000000000000000..380c8c34e7e913a1a56cecaeb72be8e0c0a8dec4
Binary files /dev/null and b/static/examples/sketch7.png differ
diff --git a/static/examples/sketch8.png b/static/examples/sketch8.png
new file mode 100644
index 0000000000000000000000000000000000000000..35b6a44f5a97738e45372e185950aa52f56d7da0
Binary files /dev/null and b/static/examples/sketch8.png differ
diff --git a/static/examples/sketch9.png b/static/examples/sketch9.png
new file mode 100644
index 0000000000000000000000000000000000000000..82484815779a1c7948a46e96c3efde8fe7b1fcd3
Binary files /dev/null and b/static/examples/sketch9.png differ
diff --git a/text2vid_modded.py b/text2vid_modded.py
new file mode 100644
index 0000000000000000000000000000000000000000..c450ed3650aa8f8c9e9bb2718c02894d0192fba3
--- /dev/null
+++ b/text2vid_modded.py
@@ -0,0 +1,612 @@
+from typing import Any, Callable, Dict, List, Optional, Union
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.models import AutoencoderKL, UNet3DConditionModel
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (
+    logging,
+    replace_example_docstring)
+from diffusers.pipelines.text_to_video_synthesis import TextToVideoSDPipelineOutput
+
+
+
+TAU_2 = 15
+TAU_1 = 10
+
+
+def init_attention_params(unet, num_frames, lambda_=None, bs=None):
+    
+    
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention": 
+            module.LAMBDA = lambda_
+            module.bs = bs
+            module.num_frames = num_frames
+            module.last_attn_slice_weights = 1
+        
+def init_attention_func(unet):
+    # ORIGINAL SOURCE CODE: https://github.com/huggingface/diffusers/blob/91ddd2a25b848df0fa1262d4f1cd98c7ccb87750/src/diffusers/models/attention.py#L276
+    # Updated source code: https://github.com/huggingface/diffusers/blob/50296739878f3e17b2d25d45ef626318b44440b9/src/diffusers/models/attention_processor.py#L571
+    def get_attention_scores(
+        self, query, key, attention_mask = None):
+        r"""
+        Compute the attention scores.
+
+        Args:
+            query (`torch.Tensor`): The query tensor.
+            key (`torch.Tensor`): The key tensor.
+            attention_mask (`torch.Tensor`, *optional*): The attention mask to use. If `None`, no mask is applied.
+
+        Returns:
+            `torch.Tensor`: The attention probabilities/scores.
+        """
+        
+        q_old = query.clone()
+        k_old = key.clone()
+        
+        if self.use_last_attn_slice:
+            if self.last_attn_slice is not None:
+                query_list = self.last_attn_slice[0]
+                key_list = self.last_attn_slice[1]
+                
+                if query.shape[1] == self.num_frames and query.shape == key.shape:
+                    
+                    key1 = key.clone()
+                    key1[:,:1,:key_list.shape[2]] = key_list[:,:1]
+                    
+                if q_old.shape == k_old.shape and q_old.shape[1]!=self.num_frames:
+                    
+                    batch_dim = query_list.shape[0] // self.bs
+                    all_dim = query.shape[0] // self.bs
+                    for i in range(self.bs):
+                        query[i*all_dim:(i*all_dim) + batch_dim,:query_list.shape[1],:query_list.shape[2]] = query_list[i*batch_dim:(i+1)*batch_dim]
+                    
+
+        dtype = query.dtype
+        if self.upcast_attention:
+            query = query.float()
+            key = key.float()
+            
+
+        if attention_mask is None:
+            baddbmm_input = torch.empty(
+                query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device
+            )
+            beta = 0
+        else:
+            baddbmm_input = attention_mask
+            beta = 1
+
+        
+        attention_scores = torch.baddbmm(
+            baddbmm_input,
+            query,
+            key.transpose(-1, -2),
+            beta=beta,
+            alpha=self.scale,
+        )
+
+        if query.shape[1] == self.num_frames and query.shape == key.shape and self.use_last_attn_slice:
+            attention_scores1 = torch.baddbmm(
+                baddbmm_input,
+                query,
+                key1.transpose(-1, -2),
+                beta=beta,
+                alpha=self.scale,
+            )
+            dynamic_lambda = torch.tensor([1 + self.LAMBDA * (i/50) for i in range(self.num_frames)]).to(dtype).cuda()
+            attention_scores[:,:self.num_frames,0] = attention_scores1[:,:self.num_frames,0] * dynamic_lambda
+            
+
+        del baddbmm_input
+
+        if self.upcast_softmax:
+            attention_scores = attention_scores.float()
+            
+        attention_probs = attention_scores.softmax(dim=-1)
+      
+
+        if self.use_last_attn_slice:
+            self.use_last_attn_slice = False
+
+        if self.save_last_attn_slice:
+
+            self.last_attn_slice = [
+                query,
+                key,
+                ]
+
+            self.save_last_attn_slice = False
+
+
+
+        del attention_scores
+        attention_probs = attention_probs.to(dtype)
+        
+
+        return attention_probs
+    
+   
+    for _, module in unet.named_modules():
+        module_name = type(module).__name__
+        
+        if module_name == "Attention":
+            module.last_attn_slice = None
+            module.use_last_attn_slice = False
+            module.save_last_attn_slice = False
+            module.LAMBDA = 0
+            module.get_attention_scores = get_attention_scores.__get__(module, type(module))
+            
+            module.bs = 0
+            module.num_frames = None
+    
+    return unet
+            
+
+def use_last_self_attention(unet, use=True):
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention" and "attn1" in name:
+            module.use_last_attn_slice = use
+            
+def save_last_self_attention(unet, save=True):
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention" and "attn1" in name:
+            module.save_last_attn_slice = save
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import TextToVideoSDPipeline
+        >>> from diffusers.utils import export_to_video
+
+        >>> pipe = TextToVideoSDPipeline.from_pretrained(
+        ...     "damo-vilab/text-to-video-ms-1.7b", torch_dtype=torch.float16, variant="fp16"
+        ... )
+        >>> pipe.enable_model_cpu_offload()
+
+        >>> prompt = "Spiderman is surfing"
+        >>> video_frames = pipe(prompt).frames[0]
+        >>> video_path = export_to_video(video_frames)
+        >>> video_path
+        ```
+"""
+
+
+# Copied from diffusers.pipelines.animatediff.pipeline_animatediff.tensor2vid
+def tensor2vid(video: torch.Tensor, processor: "VaeImageProcessor", output_type: str = "np"):
+    batch_size, channels, num_frames, height, width = video.shape
+    outputs = []
+    for batch_idx in range(batch_size):
+        batch_vid = video[batch_idx].permute(1, 0, 2, 3)
+        batch_output = processor.postprocess(batch_vid, output_type)
+
+        outputs.append(batch_output)
+
+    if output_type == "np":
+        outputs = np.stack(outputs)
+
+    elif output_type == "pt":
+        outputs = torch.stack(outputs)
+
+    elif not output_type == "pil":
+        raise ValueError(f"{output_type} does not exist. Please choose one of ['np', 'pt', 'pil']")
+
+    return outputs
+
+from diffusers import TextToVideoSDPipeline
+class TextToVideoSDPipelineModded(TextToVideoSDPipeline):
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet3DConditionModel,
+        scheduler: KarrasDiffusionSchedulers,
+    ):
+        super().__init__(vae, text_encoder, tokenizer, unet, scheduler)
+        
+
+    def call_network(self,
+                    negative_prompt_embeds,
+                    prompt_embeds,
+                    latents,
+                    inv_latents,
+                    t,
+                    i,
+                    null_embeds,
+                    cross_attention_kwargs,
+                    extra_step_kwargs,
+                    do_classifier_free_guidance,
+                    guidance_scale,
+                    ):
+         
+
+        inv_latent_model_input = inv_latents
+        inv_latent_model_input = self.scheduler.scale_model_input(inv_latent_model_input, t)
+        
+        latent_model_input = latents
+        latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+        
+
+        if do_classifier_free_guidance: 
+            noise_pred_uncond = self.unet(
+                latent_model_input,
+                t,
+                encoder_hidden_states=negative_prompt_embeds, 
+                cross_attention_kwargs=cross_attention_kwargs,
+                return_dict=False,
+            )[0]
+
+            noise_null_pred_uncond = self.unet(
+                    inv_latent_model_input,
+                    t,
+                    encoder_hidden_states=negative_prompt_embeds, 
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                    
+                    
+        if i<=TAU_2:
+            save_last_self_attention(self.unet)
+            
+
+            noise_null_pred = self.unet(
+                    inv_latent_model_input,
+                    t,
+                    encoder_hidden_states=null_embeds, 
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+            
+            if do_classifier_free_guidance:
+                noise_null_pred = noise_null_pred_uncond + guidance_scale * (noise_null_pred - noise_null_pred_uncond)
+            
+            bsz, channel, frames, width, height = inv_latents.shape
+        
+            inv_latents = inv_latents.permute(0, 2, 1, 3, 4).reshape(bsz*frames, channel, height, width)
+            noise_null_pred = noise_null_pred.permute(0, 2, 1, 3, 4).reshape(bsz*frames, channel, height, width)
+            inv_latents = self.scheduler.step(noise_null_pred, t, inv_latents, **extra_step_kwargs).prev_sample
+            inv_latents = inv_latents[None, :].reshape((bsz, frames , -1) + inv_latents.shape[2:]).permute(0, 2, 1, 3, 4)
+
+            use_last_self_attention(self.unet)
+        else:
+            noise_null_pred = None
+            
+        
+    
+        
+        noise_pred = self.unet(
+            latent_model_input,
+            t,
+            encoder_hidden_states=prompt_embeds, # For unconditional guidance
+            cross_attention_kwargs=cross_attention_kwargs,
+            return_dict=False,
+        )[0]
+
+        use_last_self_attention(self.unet, False)  
+        
+
+        if do_classifier_free_guidance:
+            noise_pred_text = noise_pred
+            noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+        # reshape latents
+        bsz, channel, frames, width, height = latents.shape
+        latents = latents.permute(0, 2, 1, 3, 4).reshape(bsz * frames, channel, width, height)
+        noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(bsz * frames, channel, width, height)
+
+        # compute the previous noisy sample x_t -> x_t-1
+        latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+        
+
+        # reshape latents back
+        latents = latents[None, :].reshape(bsz, frames, channel, width, height).permute(0, 2, 1, 3, 4)
+
+
+        return {
+            "latents": latents, 
+            "inv_latents": inv_latents,
+            "noise_pred": noise_pred,
+            "noise_null_pred": noise_null_pred,
+            }
+                
+    def optimize_latents(self, latents, inv_latents, t, i, null_embeds, cross_attention_kwargs, prompt_embeds):
+        inv_scaled = self.scheduler.scale_model_input(inv_latents, t)  
+                    
+        noise_null_pred = self.unet(
+            inv_scaled[:,:,0:1,:,:],
+            t,
+            encoder_hidden_states=null_embeds,
+            cross_attention_kwargs=cross_attention_kwargs,
+            return_dict=False,
+        )[0]
+
+        with torch.enable_grad():
+            
+            latent_train = latents[:,:,1:,:,:].clone().detach().requires_grad_(True)
+            optimizer = torch.optim.Adam([latent_train], lr=1e-3)
+
+            for j in range(10): 
+                latent_in = torch.cat([inv_latents[:,:,0:1,:,:].detach(), latent_train], dim=2)
+                latent_input_unet = self.scheduler.scale_model_input(latent_in, t)
+
+                noise_pred = self.unet(
+                    latent_input_unet,
+                    t,
+                    encoder_hidden_states=prompt_embeds, # For unconditional guidance
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+                
+                loss = torch.nn.functional.mse_loss(noise_pred[:,:,0,:,:], noise_null_pred[:,:,0,:,:])
+                
+                loss.backward()
+
+                optimizer.step()
+                optimizer.zero_grad()
+
+                print("Iteration {} Subiteration {} Loss {} ".format(i, j, loss.item()))
+            latents = latent_in.detach()
+        return latents
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_frames: int = 16,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 9.0,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        inv_latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "np",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        clip_skip: Optional[int] = None,
+        lambda_ = 0.5,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The height in pixels of the generated video.
+            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The width in pixels of the generated video.
+            num_frames (`int`, *optional*, defaults to 16):
+                The number of video frames that are generated. Defaults to 16 frames which at 8 frames per seconds
+                amounts to 2 seconds of video.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality videos at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                A higher guidance scale value encourages the model to generate images closely linked to the text
+                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`. Latents should be of shape
+                `(batch_size, num_channel, num_frames, height, width)`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"np"`):
+                The output format of the generated video. Choose between `torch.FloatTensor` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] instead
+                of a plain tuple.
+            callback (`Callable`, *optional*):
+                A function that calls every `callback_steps` steps during inference. The function is called with the
+                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function is called. If not specified, the callback is called at
+                every step.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+        Examples:
+
+        Returns:
+            [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] is
+                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
+        """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * self.vae_scale_factor
+        width = width or self.unet.config.sample_size * self.vae_scale_factor
+
+        num_images_per_prompt = 1
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds
+        )
+
+        # # 2. Define call parameters
+        # if prompt is not None and isinstance(prompt, str):
+        #     batch_size = 1
+        # elif prompt is not None and isinstance(prompt, list):
+        #     batch_size = len(prompt)
+        # else:
+        #     batch_size = prompt_embeds.shape[0]
+
+        batch_size = inv_latents.shape[0]
+        device = self._execution_device
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        text_encoder_lora_scale = (
+            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+        )
+        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+            [prompt] * batch_size,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            [negative_prompt] * batch_size if negative_prompt is not None else None,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
+            clip_skip=clip_skip,
+        )
+        null_embeds, negative_prompt_embeds = self.encode_prompt(
+            [""] * batch_size,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            [negative_prompt] * batch_size if negative_prompt is not None else None,
+            prompt_embeds=None,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
+            clip_skip=clip_skip,
+        )
+        
+        
+        
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            num_frames,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+        inv_latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            num_frames,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            inv_latents,
+        )
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+        
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        
+        init_attention_func(self.unet)
+        print("Setup for Current Run")
+        print("----------------------")
+        print("Prompt ", prompt)
+        print("Batch size ", batch_size)
+        print("Num frames ", latents.shape[2])
+        print("Lambda ", lambda_)
+        
+        init_attention_params(self.unet, num_frames=latents.shape[2], lambda_=lambda_, bs = batch_size)
+        
+        iters_to_alter = [-1]#i for i in range(0, TAU_1)]
+    
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            
+            mask_in = torch.zeros(latents.shape).to(dtype=latents.dtype, device=latents.device)
+            mask_in[:, :, 0, :, :] = 1
+            assert latents.shape[0] == inv_latents.shape[0], "Latents and Inverse Latents should have the same batch but got {} and {}".format(latents.shape[0], inv_latents.shape[0])
+            inv_latents = inv_latents.repeat(1,1,num_frames,1,1)
+
+            latents = inv_latents * mask_in + latents * (1-mask_in)
+            
+            
+
+            for i, t in enumerate(timesteps):
+                
+                curr_copy = max(1,num_frames - i)
+                inv_latents = inv_latents[:,:,:curr_copy, :, : ]
+                if i in iters_to_alter:
+
+                    latents = self.optimize_latents(latents, inv_latents, t, i, null_embeds, cross_attention_kwargs, prompt_embeds)
+                
+
+                output_dict = self.call_network(
+                        negative_prompt_embeds,
+                        prompt_embeds,
+                        latents,
+                        inv_latents,
+                        t,
+                        i,
+                        null_embeds,
+                        cross_attention_kwargs,
+                        extra_step_kwargs,
+                        do_classifier_free_guidance,
+                        guidance_scale,
+                    )
+                latents = output_dict["latents"]
+                inv_latents = output_dict["inv_latents"]
+               
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        step_idx = i // getattr(self.scheduler, "order", 1)
+                        callback(step_idx, t, latents)
+
+        # 8. Post processing
+        if output_type == "latent":
+            video = latents
+        else:
+            video_tensor = self.decode_latents(latents)
+            video = tensor2vid(video_tensor, self.image_processor, output_type)
+
+        # 9. Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return TextToVideoSDPipelineOutput(frames=video)
\ No newline at end of file
diff --git a/text2vid_modded_full.py b/text2vid_modded_full.py
new file mode 100644
index 0000000000000000000000000000000000000000..9e4aa29c674991c41e724f751030e41481e4150d
--- /dev/null
+++ b/text2vid_modded_full.py
@@ -0,0 +1,612 @@
+from typing import Any, Callable, Dict, List, Optional, Union
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTokenizer
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.models import AutoencoderKL, UNet3DConditionModel
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (
+    logging,
+    replace_example_docstring)
+from diffusers.pipelines.text_to_video_synthesis import TextToVideoSDPipelineOutput
+
+
+
+TAU_2 = 15
+TAU_1 = 10
+
+
+def init_attention_params(unet, num_frames, lambda_=None, bs=None):
+    
+    
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention": 
+            module.LAMBDA = lambda_
+            module.bs = bs
+            module.num_frames = num_frames
+            module.last_attn_slice_weights = 1
+        
+def init_attention_func(unet):
+    # ORIGINAL SOURCE CODE: https://github.com/huggingface/diffusers/blob/91ddd2a25b848df0fa1262d4f1cd98c7ccb87750/src/diffusers/models/attention.py#L276
+    # Updated source code: https://github.com/huggingface/diffusers/blob/50296739878f3e17b2d25d45ef626318b44440b9/src/diffusers/models/attention_processor.py#L571
+    def get_attention_scores(
+        self, query, key, attention_mask = None):
+        r"""
+        Compute the attention scores.
+
+        Args:
+            query (`torch.Tensor`): The query tensor.
+            key (`torch.Tensor`): The key tensor.
+            attention_mask (`torch.Tensor`, *optional*): The attention mask to use. If `None`, no mask is applied.
+
+        Returns:
+            `torch.Tensor`: The attention probabilities/scores.
+        """
+        
+        q_old = query.clone()
+        k_old = key.clone()
+        
+        if self.use_last_attn_slice:
+            if self.last_attn_slice is not None:
+                query_list = self.last_attn_slice[0]
+                key_list = self.last_attn_slice[1]
+                
+                if query.shape[1] == self.num_frames and query.shape == key.shape:
+                    
+                    key1 = key.clone()
+                    key1[:,:1,:key_list.shape[2]] = key_list[:,:1]
+                    
+                if q_old.shape == k_old.shape and q_old.shape[1]!=self.num_frames:
+                    
+                    batch_dim = query_list.shape[0] // self.bs
+                    all_dim = query.shape[0] // self.bs
+                    for i in range(self.bs):
+                        query[i*all_dim:(i*all_dim) + batch_dim,:query_list.shape[1],:query_list.shape[2]] = query_list[i*batch_dim:(i+1)*batch_dim]
+                    
+
+        dtype = query.dtype
+        if self.upcast_attention:
+            query = query.float()
+            key = key.float()
+            
+
+        if attention_mask is None:
+            baddbmm_input = torch.empty(
+                query.shape[0], query.shape[1], key.shape[1], dtype=query.dtype, device=query.device
+            )
+            beta = 0
+        else:
+            baddbmm_input = attention_mask
+            beta = 1
+
+        
+        attention_scores = torch.baddbmm(
+            baddbmm_input,
+            query,
+            key.transpose(-1, -2),
+            beta=beta,
+            alpha=self.scale,
+        )
+
+        if query.shape[1] == self.num_frames and query.shape == key.shape and self.use_last_attn_slice:
+            attention_scores1 = torch.baddbmm(
+                baddbmm_input,
+                query,
+                key1.transpose(-1, -2),
+                beta=beta,
+                alpha=self.scale,
+            )
+            dynamic_lambda = torch.tensor([1 + self.LAMBDA * (i/50) for i in range(self.num_frames)]).to(dtype).cuda()
+            attention_scores[:,:self.num_frames,0] = attention_scores1[:,:self.num_frames,0] * dynamic_lambda
+            
+
+        del baddbmm_input
+
+        if self.upcast_softmax:
+            attention_scores = attention_scores.float()
+            
+        attention_probs = attention_scores.softmax(dim=-1)
+      
+
+        if self.use_last_attn_slice:
+            self.use_last_attn_slice = False
+
+        if self.save_last_attn_slice:
+
+            self.last_attn_slice = [
+                query,
+                key,
+                ]
+
+            self.save_last_attn_slice = False
+
+
+
+        del attention_scores
+        attention_probs = attention_probs.to(dtype)
+        
+
+        return attention_probs
+    
+   
+    for _, module in unet.named_modules():
+        module_name = type(module).__name__
+        
+        if module_name == "Attention":
+            module.last_attn_slice = None
+            module.use_last_attn_slice = False
+            module.save_last_attn_slice = False
+            module.LAMBDA = 0
+            module.get_attention_scores = get_attention_scores.__get__(module, type(module))
+            
+            module.bs = 0
+            module.num_frames = None
+    
+    return unet
+            
+
+def use_last_self_attention(unet, use=True):
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention" and "attn1" in name:
+            module.use_last_attn_slice = use
+            
+def save_last_self_attention(unet, save=True):
+    for name, module in unet.named_modules():
+        module_name = type(module).__name__
+        if module_name == "Attention" and "attn1" in name:
+            module.save_last_attn_slice = save
+
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+    Examples:
+        ```py
+        >>> import torch
+        >>> from diffusers import TextToVideoSDPipeline
+        >>> from diffusers.utils import export_to_video
+
+        >>> pipe = TextToVideoSDPipeline.from_pretrained(
+        ...     "damo-vilab/text-to-video-ms-1.7b", torch_dtype=torch.float16, variant="fp16"
+        ... )
+        >>> pipe.enable_model_cpu_offload()
+
+        >>> prompt = "Spiderman is surfing"
+        >>> video_frames = pipe(prompt).frames[0]
+        >>> video_path = export_to_video(video_frames)
+        >>> video_path
+        ```
+"""
+
+
+# Copied from diffusers.pipelines.animatediff.pipeline_animatediff.tensor2vid
+def tensor2vid(video: torch.Tensor, processor: "VaeImageProcessor", output_type: str = "np"):
+    batch_size, channels, num_frames, height, width = video.shape
+    outputs = []
+    for batch_idx in range(batch_size):
+        batch_vid = video[batch_idx].permute(1, 0, 2, 3)
+        batch_output = processor.postprocess(batch_vid, output_type)
+
+        outputs.append(batch_output)
+
+    if output_type == "np":
+        outputs = np.stack(outputs)
+
+    elif output_type == "pt":
+        outputs = torch.stack(outputs)
+
+    elif not output_type == "pil":
+        raise ValueError(f"{output_type} does not exist. Please choose one of ['np', 'pt', 'pil']")
+
+    return outputs
+
+from diffusers import TextToVideoSDPipeline
+class TextToVideoSDPipelineModded(TextToVideoSDPipeline):
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet3DConditionModel,
+        scheduler: KarrasDiffusionSchedulers,
+    ):
+        super().__init__(vae, text_encoder, tokenizer, unet, scheduler)
+        
+
+    def call_network(self,
+                    negative_prompt_embeds,
+                    prompt_embeds,
+                    latents,
+                    inv_latents,
+                    t,
+                    i,
+                    null_embeds,
+                    cross_attention_kwargs,
+                    extra_step_kwargs,
+                    do_classifier_free_guidance,
+                    guidance_scale,
+                    ):
+         
+
+        inv_latent_model_input = inv_latents
+        inv_latent_model_input = self.scheduler.scale_model_input(inv_latent_model_input, t)
+        
+        latent_model_input = latents
+        latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+        
+
+        if do_classifier_free_guidance: 
+            noise_pred_uncond = self.unet(
+                latent_model_input,
+                t,
+                encoder_hidden_states=negative_prompt_embeds, 
+                cross_attention_kwargs=cross_attention_kwargs,
+                return_dict=False,
+            )[0]
+
+            noise_null_pred_uncond = self.unet(
+                    inv_latent_model_input,
+                    t,
+                    encoder_hidden_states=negative_prompt_embeds, 
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+
+                    
+                    
+        if i<=TAU_2:
+            save_last_self_attention(self.unet)
+            
+
+            noise_null_pred = self.unet(
+                    inv_latent_model_input,
+                    t,
+                    encoder_hidden_states=null_embeds, 
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+            
+            if do_classifier_free_guidance:
+                noise_null_pred = noise_null_pred_uncond + guidance_scale * (noise_null_pred - noise_null_pred_uncond)
+            
+            bsz, channel, frames, width, height = inv_latents.shape
+        
+            inv_latents = inv_latents.permute(0, 2, 1, 3, 4).reshape(bsz*frames, channel, height, width)
+            noise_null_pred = noise_null_pred.permute(0, 2, 1, 3, 4).reshape(bsz*frames, channel, height, width)
+            inv_latents = self.scheduler.step(noise_null_pred, t, inv_latents, **extra_step_kwargs).prev_sample
+            inv_latents = inv_latents[None, :].reshape((bsz, frames , -1) + inv_latents.shape[2:]).permute(0, 2, 1, 3, 4)
+
+            use_last_self_attention(self.unet)
+        else:
+            noise_null_pred = None
+            
+        
+    
+        
+        noise_pred = self.unet(
+            latent_model_input,
+            t,
+            encoder_hidden_states=prompt_embeds, # For unconditional guidance
+            cross_attention_kwargs=cross_attention_kwargs,
+            return_dict=False,
+        )[0]
+
+        use_last_self_attention(self.unet, False)  
+        
+
+        if do_classifier_free_guidance:
+            noise_pred_text = noise_pred
+            noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+        # reshape latents
+        bsz, channel, frames, width, height = latents.shape
+        latents = latents.permute(0, 2, 1, 3, 4).reshape(bsz * frames, channel, width, height)
+        noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(bsz * frames, channel, width, height)
+
+        # compute the previous noisy sample x_t -> x_t-1
+        latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
+
+        
+
+        # reshape latents back
+        latents = latents[None, :].reshape(bsz, frames, channel, width, height).permute(0, 2, 1, 3, 4)
+
+
+        return {
+            "latents": latents, 
+            "inv_latents": inv_latents,
+            "noise_pred": noise_pred,
+            "noise_null_pred": noise_null_pred,
+            }
+                
+    def optimize_latents(self, latents, inv_latents, t, i, null_embeds, cross_attention_kwargs, prompt_embeds):
+        inv_scaled = self.scheduler.scale_model_input(inv_latents, t)  
+                    
+        noise_null_pred = self.unet(
+            inv_scaled[:,:,0:1,:,:],
+            t,
+            encoder_hidden_states=null_embeds,
+            cross_attention_kwargs=cross_attention_kwargs,
+            return_dict=False,
+        )[0]
+
+        with torch.enable_grad():
+            
+            latent_train = latents[:,:,1:,:,:].clone().detach().requires_grad_(True)
+            optimizer = torch.optim.Adam([latent_train], lr=1e-3)
+
+            for j in range(10): 
+                latent_in = torch.cat([inv_latents[:,:,0:1,:,:].detach(), latent_train], dim=2)
+                latent_input_unet = self.scheduler.scale_model_input(latent_in, t)
+
+                noise_pred = self.unet(
+                    latent_input_unet,
+                    t,
+                    encoder_hidden_states=prompt_embeds, # For unconditional guidance
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    return_dict=False,
+                )[0]
+                
+                loss = torch.nn.functional.mse_loss(noise_pred[:,:,0,:,:], noise_null_pred[:,:,0,:,:])
+                
+                loss.backward()
+
+                optimizer.step()
+                optimizer.zero_grad()
+
+                print("Iteration {} Subiteration {} Loss {} ".format(i, j, loss.item()))
+            latents = latent_in.detach()
+        return latents
+
+    @torch.no_grad()
+    @replace_example_docstring(EXAMPLE_DOC_STRING)
+    def __call__(
+        self,
+        prompt: Union[str, List[str]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_frames: int = 16,
+        num_inference_steps: int = 50,
+        guidance_scale: float = 9.0,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        inv_latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "np",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        clip_skip: Optional[int] = None,
+        lambda_ = 0.5,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The height in pixels of the generated video.
+            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The width in pixels of the generated video.
+            num_frames (`int`, *optional*, defaults to 16):
+                The number of video frames that are generated. Defaults to 16 frames which at 8 frames per seconds
+                amounts to 2 seconds of video.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality videos at the
+                expense of slower inference.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                A higher guidance scale value encourages the model to generate images closely linked to the text
+                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`. Latents should be of shape
+                `(batch_size, num_channel, num_frames, height, width)`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"np"`):
+                The output format of the generated video. Choose between `torch.FloatTensor` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] instead
+                of a plain tuple.
+            callback (`Callable`, *optional*):
+                A function that calls every `callback_steps` steps during inference. The function is called with the
+                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function is called. If not specified, the callback is called at
+                every step.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+        Examples:
+
+        Returns:
+            [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.text_to_video_synthesis.TextToVideoSDPipelineOutput`] is
+                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
+        """
+        # 0. Default height and width to unet
+        height = height or self.unet.config.sample_size * self.vae_scale_factor
+        width = width or self.unet.config.sample_size * self.vae_scale_factor
+
+        num_images_per_prompt = 1
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt, height, width, callback_steps, negative_prompt, prompt_embeds, negative_prompt_embeds
+        )
+
+        # # 2. Define call parameters
+        # if prompt is not None and isinstance(prompt, str):
+        #     batch_size = 1
+        # elif prompt is not None and isinstance(prompt, list):
+        #     batch_size = len(prompt)
+        # else:
+        #     batch_size = prompt_embeds.shape[0]
+
+        batch_size = inv_latents.shape[0]
+        device = self._execution_device
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        text_encoder_lora_scale = (
+            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+        )
+        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+            [prompt] * batch_size,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            [negative_prompt] * batch_size if negative_prompt is not None else None,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
+            clip_skip=clip_skip,
+        )
+        null_embeds, negative_prompt_embeds = self.encode_prompt(
+            [""] * batch_size,
+            device,
+            num_images_per_prompt,
+            do_classifier_free_guidance,
+            [negative_prompt] * batch_size if negative_prompt is not None else None,
+            prompt_embeds=None,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
+            clip_skip=clip_skip,
+        )
+        
+        
+        
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            num_frames,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+        inv_latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            num_frames,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            inv_latents,
+        )
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+        
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        
+        init_attention_func(self.unet)
+        print("Setup for Current Run")
+        print("----------------------")
+        print("Prompt ", prompt)
+        print("Batch size ", batch_size)
+        print("Num frames ", latents.shape[2])
+        print("Lambda ", lambda_)
+        
+        init_attention_params(self.unet, num_frames=latents.shape[2], lambda_=lambda_, bs = batch_size)
+        
+        iters_to_alter = [i for i in range(0, TAU_1)]
+    
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            
+            mask_in = torch.zeros(latents.shape).to(dtype=latents.dtype, device=latents.device)
+            mask_in[:, :, 0, :, :] = 1
+            assert latents.shape[0] == inv_latents.shape[0], "Latents and Inverse Latents should have the same batch but got {} and {}".format(latents.shape[0], inv_latents.shape[0])
+            inv_latents = inv_latents.repeat(1,1,num_frames,1,1)
+
+            latents = inv_latents * mask_in + latents * (1-mask_in)
+            
+            
+
+            for i, t in enumerate(timesteps):
+                
+                curr_copy = max(1,num_frames - i)
+                inv_latents = inv_latents[:,:,:curr_copy, :, : ]
+                if i in iters_to_alter:
+
+                    latents = self.optimize_latents(latents, inv_latents, t, i, null_embeds, cross_attention_kwargs, prompt_embeds)
+                
+
+                output_dict = self.call_network(
+                        negative_prompt_embeds,
+                        prompt_embeds,
+                        latents,
+                        inv_latents,
+                        t,
+                        i,
+                        null_embeds,
+                        cross_attention_kwargs,
+                        extra_step_kwargs,
+                        do_classifier_free_guidance,
+                        guidance_scale,
+                    )
+                latents = output_dict["latents"]
+                inv_latents = output_dict["inv_latents"]
+               
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        step_idx = i // getattr(self.scheduler, "order", 1)
+                        callback(step_idx, t, latents)
+
+        # 8. Post processing
+        if output_type == "latent":
+            video = latents
+        else:
+            video_tensor = self.decode_latents(latents)
+            video = tensor2vid(video_tensor, self.image_processor, output_type)
+
+        # 9. Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (video,)
+
+        return TextToVideoSDPipelineOutput(frames=video)
\ No newline at end of file