app.py

import gradio as gr
import os
from gradio_client import Client, handle_file
import numpy as np 
import tempfile
import imageio

import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler

pipe = DiffusionPipeline.from_pretrained("cerspense/zeroscope_v2_576w", torch_dtype=torch.float16)
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_model_cpu_offload()

hf_token = os.environ.get("HF_TOKEN")

def get_caption(image_in):
    kosmos2_client = Client("fffiloni/Kosmos-2-API", hf_token=hf_token)
    kosmos2_result = kosmos2_client.predict(
		image_input=handle_file(image_in),
		text_input="Detailed",
		api_name="/generate_predictions"
    )
    print(f"KOSMOS2 RETURNS: {kosmos2_result}")

    data = kosmos2_result[1]

    # Extract and combine tokens starting from the second element
    sentence = ''.join(item['token'] for item in data[1:])

    # Find the last occurrence of "."
    #last_period_index = full_sentence.rfind('.')

    # Truncate the string up to the last period
    #truncated_caption = full_sentence[:last_period_index + 1]

    # print(truncated_caption)
    #print(f"\n—\nIMAGE CAPTION: {truncated_caption}")
    
    return sentence

def export_to_video(frames: np.ndarray, fps: int) -> str:
    frames = np.clip((frames * 255), 0, 255).astype(np.uint8)
    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 infer(image_init, progress=gr.Progress(track_tqdm=True)):
    prompt = get_caption(image_init)
    video_frames = pipe(prompt, num_inference_steps=40, height=320, width=576, num_frames=24).frames[0]
    video_path = export_to_video(video_frames, 12)
    print(video_path)
    return prompt, video_path

css = """
#col-container {max-width: 510px; margin-left: auto; margin-right: auto;}
a {text-decoration-line: underline; font-weight: 600;}
.animate-spin {
  animation: spin 1s linear infinite;
}

@keyframes spin {
  from {
      transform: rotate(0deg);
  }
  to {
      transform: rotate(360deg);
  }
}

#share-btn-container {
  display: flex; 
  padding-left: 0.5rem !important; 
  padding-right: 0.5rem !important; 
  background-color: #000000; 
  justify-content: center; 
  align-items: center; 
  border-radius: 9999px !important; 
  max-width: 13rem;
}

#share-btn-container:hover {
  background-color: #060606;
}

#share-btn {
  all: initial; 
  color: #ffffff;
  font-weight: 600; 
  cursor:pointer; 
  font-family: 'IBM Plex Sans', sans-serif; 
  margin-left: 0.5rem !important; 
  padding-top: 0.5rem !important; 
  padding-bottom: 0.5rem !important;
  right:0;
}

#share-btn * {
  all: unset;
}

#share-btn-container div:nth-child(-n+2){
  width: auto !important;
  min-height: 0px !important;
}

#share-btn-container .wrap {
  display: none !important;
}

#share-btn-container.hidden {
  display: none!important;
}
img[src*='#center'] { 
    display: block;
    margin: auto;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(
            """
            <h1 style="text-align: center;">Zeroscope Image-to-Video</h1>
            <p style="text-align: center;">
            A watermark-free Modelscope-based video model optimized for producing high-quality 16:9 compositions and a smooth video output. <br />
            This demo is a variation that lets you upload an image as reference for video generation. 
            </p>
            
            [![Duplicate this Space](https://huggingface.co/datasets/huggingface/badges/raw/main/duplicate-this-space-sm.svg#center)](https://huggingface.co/spaces/fffiloni/zeroscope-img-to-video?duplicate=true)
            
            """
        )

        image_init = gr.Image(label="Image Init", type="filepath", sources=["upload"], elem_id="image-init")
        #inference_steps = gr.Slider(label="Inference Steps", minimum=10, maximum=100, step=1, value=40, interactive=False)
        submit_btn = gr.Button("Submit")
        coca_cap = gr.Textbox(label="Caption", placeholder="Kosmos-2 caption will be displayed here", elem_id="coca-cap-in")
        video_result = gr.Video(label="Video Output", elem_id="video-output")

    submit_btn.click(
        fn=infer,
        inputs=[image_init],
        outputs=[coca_cap, video_result],
        show_api=False
    )

demo.queue(max_size=12).launch(show_api=False)