app.py

import gradio as gr
import os 
hf_token = os.environ.get('HF_TOKEN')

lpmc_client = gr.load("seungheondoh/LP-Music-Caps-demo", src="spaces")

from gradio_client import Client

client = Client("https://fffiloni-test-llama-api-debug.hf.space/", hf_token=hf_token)
zrscp_client = Client("https://fffiloni-zeroscope--rhhd9.hf.space/", hf_token=hf_token)

from pydub import AudioSegment

def cut_audio(input_path, output_path, max_duration=30000):
    audio = AudioSegment.from_file(input_path)

    if len(audio) > max_duration:
        audio = audio[:max_duration]

    audio.export(output_path, format="mp3")

    return output_path

def solo_zrscp(prompt):
    res_vid = zrscp_client.predict(
        prompt,
        api_name="/zrscp"
    )
    return res_vid

def infer(audio_file):

    truncated_audio = cut_audio(audio_file, "trunc_audio.mp3")
    
    cap_result = lpmc_client(
    				truncated_audio,	# str (filepath or URL to file) in 'audio_path' Audio component
    				api_name="predict"
    )
    print(cap_result)

    #summarize_q = f"""

    #I'll give you a list of music descriptions. Create a summary reflecting the musical ambiance. 
    #Do not processs each segment, but provide a summary for the whole instead.
    
    #Here's the list:

    #{cap_result}
    #"""

    #summary_result = client.predict(
    #				summarize_q,	# str in 'Message' Textbox component
    #				api_name="/chat_1"
    #)

    #print(f"SUMMARY: {summary_result}")

    llama_q = f"""
    I'll give you a music description. 
    Give me an image description that would fit well with the music description. 
    Be creative, do not do list, just an image description as required. Try to think about human characters first.
    Your image description must fit well for a stable diffusion prompt.
    
    Here's the music description :
    
    « {cap_result} »
    """
    
    result = client.predict(
    				llama_q,	# str in 'Message' Textbox component
                    "M2I",
    				api_name="/predict"
    )

    
    
    
    print(f"Llama2 result: {result}")

    res_vid = zrscp_client.predict(
        result,
        api_name="/zrscp"
    )

    print("Finished")
    
    #return cap_result, result, images
    return res_vid, result, gr.update(visible=True)

css = """
#col-container {max-width: 510px; margin-left: auto; margin-right: auto;}
"""
with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.HTML("""<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;">
                    Music To Zeroscope Video
                </h1>
                </div>
                <p style="margin-bottom: 10px; font-size: 94%">
                Sends an audio into <a href="https://huggingface.co/spaces/seungheondoh/LP-Music-Caps-demo" target="_blank">LP-Music-Caps</a>
                to generate a audio caption which is then translated to an illustrative image description with Llama2, and finally run through 
                Zeroscope to generate a 3s video from the audio ! <br /><br />
                Note: Only the first 30 seconds of your audio will be used for inference.
                </p>
            </div>""")
        audio_input = gr.Audio(label="Music input", type="filepath", source="upload")
        infer_btn = gr.Button("Generate Video from Music")
        #lpmc_cap = gr.Textbox(label="Lp Music Caps caption")
        llama_trans_cap = gr.Textbox(label="Llama translation", visible=False)
        vid_result = gr.Video(label="Image Result")
        tryagain_btn = gr.Button("Try again ?", visible=False)

        gr.Examples(examples=[["./examples/electronic.mp3"],["./examples/folk.wav"], ["./examples/orchestra.wav"]],
                    fn=infer,
                    inputs=[audio_input],
                    outputs=[vid_result, llama_trans_cap, tryagain_btn],
                    cache_examples=True
                   )

    #infer_btn.click(fn=infer, inputs=[audio_input], outputs=[lpmc_cap, llama_trans_cap, img_result])
    infer_btn.click(fn=infer, inputs=[audio_input], outputs=[vid_result, llama_trans_cap, tryagain_btn])
    tryagain_btn.click(fn=solo_zrscp, inputs=[llama_trans_cap], outputs=[vid_result])

demo.queue(max_size=20).launch()