motion_utils/video_generator_animatediff.py

import torch
import numpy as np
import os
import imageio
from PIL import Image
from diffusers import AnimateDiffPipeline
from diffusers.utils import load_image

def generate_video_from_image(image_path, prompt, duration, fps, output_path):
    # Charger l’image et la convertir en RGB
    image = load_image(image_path).convert("RGB")

    # Utiliser un modèle AnimateDiff compatible (UNetMotionModel)
    model_id = "stabilityai/stable-video-diffusion-img2vid"
    pipe = AnimateDiffPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
    pipe.enable_model_cpu_offload()

    # Calculer le nombre de frames à générer
    num_frames = duration * fps

    # Générer les frames à partir du prompt et de l’image
    output = pipe(prompt=prompt, image=image, num_inference_steps=25, guidance_scale=7.5, num_frames=num_frames)
    frames = output.frames

    # Convertir les frames en tableau numpy et sauvegarder en vidéo
    frames = [np.array(frame) for frame in frames]
    imageio.mimwrite(output_path, frames, fps=fps)