app.py

#!/usr/bin/env python

from __future__ import annotations

import os
import random
import tempfile

import gradio as gr
import imageio
import numpy as np
import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler

DESCRIPTION = '# zeroscope v2'
if not torch.cuda.is_available():
    DESCRIPTION += '\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>'
if (SPACE_ID := os.getenv('SPACE_ID')) is not None:
    DESCRIPTION += f'\n<p>For faster inference without waiting in queue, you may duplicate the space and upgrade to GPU in settings. <a href="https://huggingface.co/spaces/{SPACE_ID}?duplicate=true"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://bit.ly/3gLdBN6" alt="Duplicate Space" /></a></p>'

MAX_NUM_FRAMES = int(os.getenv('MAX_NUM_FRAMES', '200'))
DEFAULT_NUM_FRAMES = min(MAX_NUM_FRAMES,
                         int(os.getenv('DEFAULT_NUM_FRAMES', '24')))
MAX_SEED = np.iinfo(np.int32).max

if torch.cuda.is_available():
    pipe = DiffusionPipeline.from_pretrained('cerspense/zeroscope_v2_576w',
                                             torch_dtype=torch.float16)
    pipe.enable_model_cpu_offload()
else:
    pipe = DiffusionPipeline.from_pretrained('cerspense/zeroscope_v2_576w')
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_vae_slicing()


def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed


def to_video(frames: list[np.ndarray], fps: int) -> str:
    out_file = tempfile.NamedTemporaryFile(suffix='.mp4', delete=False)
    writer = imageio.get_writer(out_file.name, format='FFMPEG', fps=fps)
    for frame in frames:
        writer.append_data(frame)
    writer.close()
    return out_file.name


def generate(prompt: str, seed: int, num_frames: int,
             num_inference_steps: int) -> str:
    generator = torch.Generator().manual_seed(seed)
    frames = pipe(prompt,
                  num_inference_steps=num_inference_steps,
                  num_frames=num_frames,
                  width=576,
                  height=320,
                  generator=generator).frames
    return to_video(frames, 8)


examples = [
    ['An astronaut riding a horse', 0, 24, 25],
    ['A panda eating bamboo on a rock', 0, 24, 25],
    ['Spiderman is surfing', 0, 24, 25],
]

with gr.Blocks(css='style.css') as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Box():
        with gr.Row():
            prompt = gr.Text(label='Prompt',
                             show_label=False,
                             max_lines=1,
                             placeholder='Enter your prompt',
                             container=False)
            run_button = gr.Button('Generate video', scale=0)
        result = gr.Video(label='Result', show_label=False)
        with gr.Accordion('Advanced options', open=False):
            seed = gr.Slider(label='Seed',
                             minimum=0,
                             maximum=MAX_SEED,
                             step=1,
                             value=0)
            randomize_seed = gr.Checkbox(label='Randomize seed', value=True)
            num_frames = gr.Slider(
                label='Number of frames',
                minimum=24,
                maximum=MAX_NUM_FRAMES,
                step=1,
                value=24,
                info=
                'Note that the content of the video also changes when you change the number of frames.'
            )
            num_inference_steps = gr.Slider(label='Number of inference steps',
                                            minimum=10,
                                            maximum=50,
                                            step=1,
                                            value=25)

    inputs = [
        prompt,
        seed,
        num_frames,
        num_inference_steps,
    ]
    gr.Examples(examples=examples,
                inputs=inputs,
                outputs=result,
                fn=generate,
                cache_examples=os.getenv('CACHE_EXAMPLES') == '1')

    prompt.submit(
        fn=randomize_seed_fn,
        inputs=[seed, randomize_seed],
        outputs=seed,
        queue=False,
    ).then(
        fn=generate,
        inputs=inputs,
        outputs=result,
        api_name='run',
    )
    run_button.click(
        fn=randomize_seed_fn,
        inputs=[seed, randomize_seed],
        outputs=seed,
        queue=False,
    ).then(
        fn=generate,
        inputs=inputs,
        outputs=result,
    )
demo.queue(max_size=10).launch()