app.py

import gradio as gr
import os
import time
from moviepy.editor import *
import imageio
#from share_btn import community_icon_html, loading_icon_html, share_js

#token = os.environ.get('HF_TOKEN')
caption = gr.Blocks.load(name="spaces/fffiloni/CoCa-clone")
audio_gen = gr.Blocks.load(name="spaces/fffiloni/audioldm-text-to-audio-generation-copy")

ph_message="If you're not happy with sound result, you can manually describe the scene depicted in your video, with audioLDM recommendations in mind :)"

def extract_video_frames(video_in):

    # Load the video file
    clip = VideoFileClip(video_in)
    
    # Calculate the total duration of the video
    total_duration = clip.duration

    if total_duration > 5:
        clip = clip.subclip(0,5)
        total_duration = clip.duration
    
    # Set the intervals to extract the frames
    intervals = [0, 2, 4, total_duration]
    
    # Initialize the list to store the extracted frames
    frames = []
    
    # Iterate through the intervals and extract the frames
    for i, interval in enumerate(intervals):
        # Get the frame at the given time
        frame = clip.get_frame(interval)

        # Save the frame as JPG image
        imageio.imwrite(f'frame{i}.jpg', frame)
        # Add the frame to the list
        frames.append(f'frame{i}.jpg')
    print(frames)
    return frames
    

def input_changes(input_vid):
    
    if input_vid == None:
        return manual_cap.update(value="",placeholder=ph_message), caption_output.update(value=None), video_output.update(value=None), sound_output.update(value=None)
    else:
        picked_frames = extract_video_frames(input_vid)
        caps = []
        for one_frame in picked_frames:
            cap = caption(one_frame, "Nucleus sampling", 1.2, 0.5, 5, 20, fn_index=0)
            caps.append(f"the sound of {cap}")
            print(caps)
        final_cap = '\n then '.join(caps)
        print(final_cap)
        print("CoCa caption: '" + final_cap + "' • ")
        ph_update = "CoCa caption: '" + final_cap + "' • "
        
        return manual_cap.update(value="",placeholder=f"{ph_update}{ph_message}"), caption_output.update(value=final_cap), video_output.update(value=None), sound_output.update(value=None)
 
def infer(video_input, manual_caption, duration_in, seed, caption_output):
    
    print(duration_in)
    if manual_caption == "":
        cap = caption_output
        #cap = caption(image_input, fn_index=0)
        #print("CoCa caption: '" + cap + "' • ")
        #ph_update = "CoCa caption: '" + cap + "' • "
    else:
        cap = manual_caption
        print("manual caption: " + cap)
        ph_update=""
    
    sound = audio_gen(cap, duration_in, 2.5, seed, 3, "audioldm-m-text-ft", fn_index=0)

    print(sound)
  
    # AudioLDM loaded demo returns a video, so we only keep the audio
    video = VideoFileClip(sound)
    audio = video.audio
    audio.write_audiofile("sound.mp3")

    # Then we put the audio to the original video
    
    # Load the input video file
    video_in = VideoFileClip(video_input)
    duration = video_in.duration
    if duration > 5:
        video_in = video_in.subclip(0,5)

    new_audio = AudioFileClip("sound.mp3")
    # Make the audio the same length as the video
    new_audio = new_audio.set_duration(video_in.duration)
    
    # Combine the audio and video
    result = video_in.set_audio(new_audio)
    
    # Save the result
    result.write_videofile("result.mp4", codec='libx264', audio_codec='aac')

    
    #return cap, sound[1], gr.Textbox.update(placeholder=f"{ph_update}{ph_message}"), gr.Group.update(visible=True)
    #return cap, "sound.mp3", gr.Group.update(visible=True)
    return cap, "result.mp4", "sound.mp3"

title = """
    <div style="text-align: center; max-width: 700px; margin: 0 auto;">
        <div
        style="
            display: inline-flex;
            align-items: center;
            gap: 0.8rem;
            font-size: 1.75rem;
        "
        >
        <h1 style="font-weight: 900; margin-bottom: 7px; margin-top: 5px;">
            Video to Sound Effect
        </h1>
        </div>
        <p style="margin-bottom: 10px; font-size: 94%">
        Convert images from video to a corresponding sound effect generated through CoCa Image Captioning & AudioLDM. <br />
        This demo is experimental and works only with exactly 5 seconds videos.
        </p>
    </div>
"""

article = """
    
    <div class="footer">
        <p>
         
        Follow <a href="https://twitter.com/fffiloni" target="_blank">Sylvain Filoni</a> for future updates 🤗
        </p>
    </div>

    <div id="may-like-container" style="display: flex;justify-content: center;flex-direction: column;align-items: center;margin-bottom: 30px;">
        <p>You may also like: </p>
        
        <div id="may-like-content" style="display:flex;flex-wrap: wrap;align-items:center;height:20px;">
            
            <svg height="20" width="208" style="margin-left:4px;margin-bottom: 6px;">       
                 <a href="https://huggingface.co/spaces/haoheliu/audioldm-text-to-audio-generation" target="_blank">
                    <image href="https://img.shields.io/badge/🤗 Spaces-AudioLDM_Text_to_Audio-blue" src="https://img.shields.io/badge/🤗 Spaces-AudioLDM_Text_to_Audio-blue.png" height="20"/>
                 </a>
            </svg>

            <svg height="20" width="122" style="margin-left:4px;margin-bottom: 6px;">       
                 <a href="https://huggingface.co/spaces/fffiloni/spectrogram-to-music" target="_blank">
                    <image href="https://img.shields.io/badge/🤗 Spaces-Riffusion-blue" src="https://img.shields.io/badge/🤗 Spaces-Riffusion-blue.png" height="20"/>
                 </a>
            </svg>
            
        </div>
    </div>
"""

with gr.Blocks(css="style.css") as demo:
    with gr.Column(elem_id="col-container"):
        
        gr.HTML(title)
    
        input_vid = gr.Video(source="upload", type="filepath", elem_id="input-vid")
        
        with gr.Column():
            manual_cap = gr.Textbox(label="Manual Video description (optional)", lines=3, placeholder=ph_message)
            with gr.Row():
                duration_in = gr.Slider(interactive=False, minimum=5, maximum=10, step=5, value=5, label="Duration")
                seed_in = gr.Slider(label="Seed", value=440, minimum=45, maximum=10000, step=1)
        
        caption_output = gr.Textbox(label="Caption", visible=False, elem_id="text-caption")
        video_output = gr.Video(label="Result", elem_id="video-output")
        sound_output = gr.Audio()
        #debug = gr.Textbox()
        generate = gr.Button("Generate SFX from Video")

        #with gr.Group(elem_id="share-btn-container", visible=False) as share_group:
        #    community_icon = gr.HTML(community_icon_html)
        #    loading_icon = gr.HTML(loading_icon_html)
        #    share_button = gr.Button("Share to community", elem_id="share-btn")

        gr.HTML(article)

    change_out = [manual_cap, caption_output, video_output, sound_output]
    input_vid.change(input_changes, input_vid, change_out, queue=False)
    

    
    generate.click(infer, inputs=[input_vid, manual_cap, duration_in, seed_in, caption_output], outputs=[caption_output, video_output, sound_output], api_name="v2fx")
    #share_button.click(None, [], [], _js=share_js)

demo.queue(max_size=32).launch(debug=True)