app.py

import gradio as gr

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

client = Client("https://ysharma-explore-llamav2-with-tgi.hf.space/")

from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, use_safetensors=True, variant="fp16")
pipe.to("cuda")

# if using torch < 2.0
# pipe.enable_xformers_memory_efficient_attention()

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 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 music description, then i want you to provide an illustrative image description that would fit well with the music.
    Answer with only one image description. Never do lists.

    Here's the music description :

    {summary_result}
    
    """

    result = client.predict(
    				llama_q,	# str in 'Message' Textbox component
    				api_name="/chat_1"
    )
    
    print(result)

    images = pipe(prompt=result).images[0]
    
    #return cap_result, result, images
    return images

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 Image
                </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 
                Stable Diffusion XL to generate an image 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(type="filepath", source="upload")
        infer_btn = gr.Button("Generate Image from Music")
        #lpmc_cap = gr.Textbox(label="Lp Music Caps caption")
        #llama_trans_cap = gr.Textbox(label="Llama translation")
        img_result = gr.Gallery(label="Result", grid=2)

    #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=[img_result])

demo.queue().launch()