app.py

import subprocess
subprocess.run(
    'pip install numpy==1.26.4',
    shell=True
)

import os
import gradio as gr
import torch
import spaces
import random
from PIL import Image
import numpy as np

from glob import glob
from pathlib import Path
from typing import Optional

from diffsynth import save_video, ModelManager, SVDVideoPipeline
from diffsynth import SDVideoPipeline, ControlNetConfigUnit, VideoData, save_frames
from diffsynth.extensions.RIFE import RIFESmoother

import uuid

HF_TOKEN = os.environ.get("HF_TOKEN", None)
# Constants
MAX_SEED = np.iinfo(np.int32).max
CSS = """
footer {
    visibility: hidden;
}
"""

JS = """function () {
  gradioURL = window.location.href
  if (!gradioURL.endsWith('?__theme=dark')) {
    window.location.replace(gradioURL + '?__theme=dark');
  }
}"""


# Ensure model and scheduler are initialized in GPU-enabled function
if torch.cuda.is_available():
    model_manager = ModelManager(
        torch_dtype=torch.float16, 
        device="cuda", 
        model_id_list=["stable-video-diffusion-img2vid-xt", "ExVideo-SVD-128f-v1"],
        downloading_priority=["HuggingFace"])
    pipe = SVDVideoPipeline.from_model_manager(model_manager)
        

    model_manager2 = ModelManager(torch_dtype=torch.float16, device="cuda")
    model_manager2.load_textual_inversions("models/textual_inversion")
    model_manager2.load_models([
            "models/stable_diffusion/flat2DAnimerge_v45Sharp.safetensors",
            "models/AnimateDiff/mm_sd_v15_v2.ckpt",
            "models/ControlNet/control_v11p_sd15_lineart.pth",
            "models/ControlNet/control_v11f1e_sd15_tile.pth",
            "models/RIFE/flownet.pkl"
    ])
    pipe2 = SDVideoPipeline.from_model_manager(
        model_manager2,
        [
            ControlNetConfigUnit(
                processor_id="lineart",
                model_path="models/ControlNet/control_v11p_sd15_lineart.pth",
                scale=0.5
            ),
            ControlNetConfigUnit(
                processor_id="tile",
                model_path="models/ControlNet/control_v11f1e_sd15_tile.pth",
                scale=0.5
            )
        ]
    )
    smoother = RIFESmoother.from_model_manager(model_manager2)


def video_to_image(selected):
    if selected == "ExVideo":
        return gr.Image(label='Upload Image', height=600, scale=2, image_mode="RGB", type="filepath")



@spaces.GPU(duration=120)
def generate(
    media,
    selected,
    seed: Optional[int] = -1,
    num_inference_steps: int = 10,
    animatediff_batch_size: int = 32,
    animatediff_stride: int = 16,
    motion_bucket_id: int = 127,
    fps_id: int = 25,
    num_frames: int = 50,
    prompt,
    output_folder: str = "outputs",
    progress=gr.Progress(track_tqdm=True)):

    print(media)
    
    if seed == -1:
        seed = random.randint(0, MAX_SEED)

    torch.manual_seed(seed)
    
    os.makedirs(output_folder, exist_ok=True)
    base_count = len(glob(os.path.join(output_folder, "*.mp4")))
    video_path = os.path.join(output_folder, f"{base_count:06d}.mp4")

    if selected == "ExVideo":
        image = Image.open(media)
        video = pipe(
            input_image=image.resize((512, 512)), 
            num_frames=num_frames, 
            fps=fps_id, 
            height=512, 
            width=512,
            motion_bucket_id=motion_bucket_id,
            num_inference_steps=num_inference_steps,
            min_cfg_scale=2, 
            max_cfg_scale=2, 
            contrast_enhance_scale=1.2
        )
        model_manager.to("cpu")
    else:
        up_video = VideoData(
            video_file=media,
            height=1024, width=1024)
        input_video = [up_video[i] for i in range(40*60, 41*60)]

        video = pipe(
            prompt=prompt,
            negative_prompt="verybadimagenegative_v1.3",
            cfg_scale=3, 
            clip_skip=2,
            controlnet_frames=input_video, num_frames=len(input_video),
            num_inference_steps=num_inference_steps, 
            height=1024, 
            width=1024,
            animatediff_batch_size=animatediff_batch_size, 
            animatediff_stride=animatediff_stride,
            vram_limit_level=0,
        )
        video = smoother(video)

    
    save_video(video, video_path, fps=fps_id)
    
    return video_path, seed


examples = [
        "./train.jpg",
        "./girl.webp",
        "./robo.jpg",
        './working.mp4',
    ]



# Gradio Interface

with gr.Blocks(css=CSS, js=JS, theme="soft") as demo:
    gr.HTML("<h1><center>Exvideo📽️Diffutoon</center></h1>")
    gr.HTML("<p><center>Exvideo and Diffutoon video generation<br><b>Update</b>: first version<br><b>Note</b>: ZeroGPU limited, Set the parameters appropriately.</center></p>")
    with gr.Row():
        media = gr.Video(label='Upload Video', height=600, scale=2)
        video = gr.Video(label="Generated Video", height=600, scale=2)
        with gr.Column(scale=1):
            selected = gr.Radio(
                label="Selected App",
                    choices=["ExVideo", "Diffutoon"],
                    value="Diffutoon"
            )
            seed = gr.Slider(
                label="Seed (-1 Random)",
                minimum=-1,
                maximum=MAX_SEED,
                step=1,
                value=-1,
                )
            num_inference_steps = gr.Slider(
                label="Inference steps", 
                info="Inference steps",
                step=1,
                value=10, 
                minimum=1, 
                maximum=50
                )
            with gr.Accordion("Diffutoon Options", open=False):
                animatediff_batch_size = gr.Slider(
                    label="Animatediff batch size",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=32,
                )
                animatediff_stride = gr.Slider(
                    label="Animatediff stride",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=16,
                ) 
            with gr.Accordion("ExVideo Options", open=False):
                motion_bucket_id = gr.Slider(
                    label="Motion bucket id", 
                    info="Controls how much motion to add/remove from the image", 
                    value=127, 
                    step=1,
                    minimum=1, 
                    maximum=255
                )
                fps_id = gr.Slider(
                    label="Frames per second", 
                    info="The length of your video in seconds will be 25/fps", 
                    value=6,
                    step=1,
                    minimum=5, 
                    maximum=30
                )
                num_frames = gr.Slider(
                    label="Frames num", 
                    info="Frames num",
                    step=1,
                    value=50, 
                    minimum=1, 
                    maximum=128
                )
    prompt = gr.Textbox(label="Prompt")
    with gr.Row():
        submit_btn = gr.Button(value="Generate")
        #stop_btn = gr.Button(value="Stop", variant="stop")
        clear_btn = gr.ClearButton([media, seed, video])
    gr.Examples(
        examples=examples,
        examples_per_page=4,
    )

    selected.change(fn=video_to_image, inputs=[selected], outputs=[media])
    submit_event = submit_btn.click(fn=generate, inputs=[media, selected, seed, num_inference_steps, animatediff_batch_size, animatediff_stride, motion_bucket_id, fps_id, num_frames, prompt], outputs=[video, seed], api_name="video")
    #stop_btn.click(fn=None, inputs=None, outputs=None, cancels=[submit_event])
    
demo.queue().launch()