Update app.py

app.py

@@ -2,28 +2,9 @@ import torch
 import gradio as gr
 from diffusers import AnimateDiffPipeline, MotionAdapter, DPMSolverMultistepScheduler, AutoencoderKL, SparseControlNetModel, EulerAncestralDiscreteScheduler
 from diffusers.utils import export_to_gif, load_image
-from realesrgan import RealESRGAN
-from PIL import Image
-import cv2
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
-def enhance_quality(image_path):
-    model = RealESRGAN(device, scale=4)
-    model.load_weights('RealESRGAN_x4.pth', download=True)
-    img = Image.open(image_path)
-    sr_image = model.predict(img)
-    enhanced_path = 'enhanced_' + image_path
-    sr_image.save(enhanced_path)
-    return enhanced_path
-
-def denoise_image(image_path):
-    image = cv2.imread(image_path)
-    denoised_image = cv2.fastNlMeansDenoisingColored(image, None, 10, 10, 7, 21)
-    denoised_path = 'denoised_' + image_path
-    cv2.imwrite(denoised_path, denoised_image)
-    return denoised_path
-
 def generate_video(prompt, negative_prompt, num_inference_steps, conditioning_frame_indices, controlnet_conditioning_scale):
     motion_adapter = MotionAdapter.from_pretrained("guoyww/animatediff-motion-adapter-v1-5-3", torch_dtype=torch.float16).to(device)
     controlnet = SparseControlNetModel.from_pretrained("guoyww/animatediff-sparsectrl-scribble", torch_dtype=torch.float16).to(device)
@@ -60,9 +41,7 @@ def generate_video(prompt, negative_prompt, num_inference_steps, conditioning_fr
     ).frames[0]
     
     export_to_gif(video, "output.gif")
-    enhanced_gif = enhance_quality("output.gif")
-    denoised_gif = denoise_image(enhanced_gif)
-    return denoised_gif
+    return "output.gif"
 
 def generate_simple_video(prompt):
     adapter = MotionAdapter.from_pretrained("guoyww/animatediff-motion-adapter-v1-5-2", torch_dtype=torch.float16).to(device)
@@ -79,23 +58,21 @@ def generate_simple_video(prompt):
 
     frames = pipe(
         prompt,
-        num_frames=128,  # Increased for smoother video
-        num_inference_steps=100,  # Increased for higher quality
-        guidance_scale=15.0,  # Increased for stronger guidance
+        num_frames=128,
+        num_inference_steps=100,
+        guidance_scale=15.0,
         decode_chunk_size=1,
     ).frames[0]
     
     export_to_gif(frames, "simple_output.gif")
-    enhanced_gif = enhance_quality("simple_output.gif")
-    denoised_gif = denoise_image(enhanced_gif)
-    return denoised_gif
+    return "simple_output.gif"
 
 demo1 = gr.Interface(
     fn=generate_video,
     inputs=[
         gr.Textbox(label="Prompt", value="an aerial view of a cyberpunk city, night time, neon lights, masterpiece, high quality"),
         gr.Textbox(label="Negative Prompt", value="low quality, worst quality, letterboxed"),
-        gr.Slider(label="Number of Inference Steps", minimum=1, maximum=200, step=1, value=100),  # Increased default value
+        gr.Slider(label="Number of Inference Steps", minimum=1, maximum=200, step=1, value=100),
         gr.Textbox(label="Conditioning Frame Indices", value="[0, 8, 15]"),
         gr.Slider(label="ControlNet Conditioning Scale", minimum=0.1, maximum=2.0, step=0.1, value=1.0)
     ],
@@ -114,5 +91,4 @@ demo2 = gr.Interface(
 
 demo = gr.TabbedInterface([demo1, demo2], ["Advanced Video Generation", "Simple Video Generation"])
 
-demo.launch()
-#demo.launch(server_name="0.0.0.0", server_port=7910)
+demo.launch()