import gradio as gr import torch import os import spaces import uuid from diffusers import AnimateDiffPipeline, MotionAdapter, EulerDiscreteScheduler from diffusers.utils import export_to_video from huggingface_hub import hf_hub_download from safetensors.torch import load_file from PIL import Image # Constants bases = { "Cartoon": "frankjoshua/toonyou_beta6", "Realistic": "emilianJR/epiCRealism", "3d": "Lykon/DreamShaper", "Anime": "Yntec/mistoonAnime2" } step_loaded = None base_loaded = "Realistic" motion_loaded = None # Ensure model and scheduler are initialized in GPU-enabled function if not torch.cuda.is_available(): raise NotImplementedError("No GPU detected!") device = "cuda" dtype = torch.float16 pipe = AnimateDiffPipeline.from_pretrained(bases[base_loaded], torch_dtype=dtype).to(device) pipe.scheduler = EulerDiscreteScheduler.from_config(pipe.scheduler.config, timestep_spacing="trailing", beta_schedule="linear") # Safety checkers from transformers import CLIPFeatureExtractor feature_extractor = CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32") # Function @spaces.GPU(duration=30,queue=False) def generate_image(prompt, base="Realistic", motion="", step=8, resolution="Square", progress=gr.Progress()): global step_loaded global base_loaded global motion_loaded print(prompt, base, step, resolution) # Set resolution if resolution == "Square": width, height = 512, 512 elif resolution == "Horizontal": width, height = 1280, 720 if step_loaded != step: repo = "ByteDance/AnimateDiff-Lightning" ckpt = f"animatediff_lightning_{step}step_diffusers.safetensors" pipe.unet.load_state_dict(load_file(hf_hub_download(repo, ckpt), device=device), strict=False) step_loaded = step if base_loaded != base: pipe.unet.load_state_dict(torch.load(hf_hub_download(bases[base], "unet/diffusion_pytorch_model.bin"), map_location=device), strict=False) base_loaded = base if motion_loaded != motion: pipe.unload_lora_weights() if motion != "": pipe.load_lora_weights(motion, adapter_name="motion") pipe.set_adapters(["motion"], [0.7]) motion_loaded = motion progress((0, step)) def progress_callback(i, t, z): progress((i+1, step)) output = pipe(prompt=prompt, guidance_scale=1.2, num_inference_steps=step, width=width, height=height, callback=progress_callback, callback_steps=1) name = str(uuid.uuid4()).replace("-", "") path = f"/tmp/{name}.mp4" export_to_video(output.frames[0], path, fps=10) return path # Gradio Interface with gr.Blocks(css="style.css") as demo: gr.HTML( "