app.py

import gradio as gr
import torch

from PIL import Image
import numpy as np
from spectro import wav_bytes_from_spectrogram_image

from diffusers import StableDiffusionPipeline
from diffusers import StableDiffusionImg2ImgPipeline

#from share_btn import community_icon_html, loading_icon_html, share_js

device = "cuda"
MODEL_ID = "riffusion/riffusion-model-v1"
pipe = StableDiffusionPipeline.from_pretrained(MODEL_ID, torch_dtype=torch.float16)
pipe = pipe.to(device)
pipe2 = StableDiffusionImg2ImgPipeline.from_pretrained(MODEL_ID, torch_dtype=torch.float16)
pipe2 = pipe2.to(device)

spectro_from_wav = gr.Interface.load("spaces/fffiloni/audio-to-spectrogram")

def predict(prompt, negative_prompt, audio_input, duration):
    if audio_input == None :
        return classic(prompt, negative_prompt, duration)
    else :
        return style_transfer(prompt, negative_prompt, audio_input)

def classic(prompt, negative_prompt, duration):
    if duration == 5:
        width_duration=512
    else :
        width_duration = 512 + ((int(duration)-5) * 128)
    spec = pipe(prompt, negative_prompt=negative_prompt, height=512, width=width_duration).images[0]
    print(spec)
    wav = wav_bytes_from_spectrogram_image(spec)
    with open("output.wav", "wb") as f:
        f.write(wav[0].getbuffer())
    return spec, 'output.wav'

def style_transfer(prompt, negative_prompt, audio_input):
    spec = spectro_from_wav(audio_input)
    print(spec)
    # Open the image
    im = Image.open(spec)
    
    
    # Open the image
    im = image_from_spectrogram(im, 1)
   
    
    new_spectro = pipe2(prompt=prompt, image=im, strength=0.5, guidance_scale=7).images
    wav = wav_bytes_from_spectrogram_image(new_spectro[0])
    with open("output.wav", "wb") as f:
        f.write(wav[0].getbuffer())
    return new_spectro[0], 'output.wav'

def image_from_spectrogram(
    spectrogram: np.ndarray, max_volume: float = 50, power_for_image: float = 0.25
) -> Image.Image:
    """
    Compute a spectrogram image from a spectrogram magnitude array.
    """
    # Apply the power curve
    data = np.power(spectrogram, power_for_image)

    # Rescale to 0-255
    data = data * 255 / max_volume

    # Invert
    data = 255 - data

    # Convert to a PIL image
    image = Image.fromarray(data.astype(np.uint8))

    # Flip Y
    image = image.transpose(Image.FLIP_TOP_BOTTOM)

    # Convert to RGB
    image = image.convert("RGB")

    return image

title = """
    <div style="text-align: center; max-width: 500px; margin: 0 auto;">
        <div
        style="
            display: inline-flex;
            align-items: center;
            gap: 0.8rem;
            font-size: 1.75rem;
            margin-bottom: 10px;
            line-height: 1em;
        "
        >
        <h1 style="font-weight: 600; margin-bottom: 7px;">
            Riffusion real-time music generation
        </h1>
        </div>
        <p style="margin-bottom: 10px;font-size: 94%;font-weight: 100;line-height: 1.5em;">
        Describe a musical prompt, generate music by getting a spectrogram image & sound.
        </p>
    </div>
"""

article = """
    <p style="text-align: center;font-size: 94%;margin-bottom: 20px;">
        Do you need faster results ? You can skip the queue by duplicating this space: 
        <span style="display: flex;align-items: center;justify-content: center;height: 30px;">
            <a style="margin-right: 10px;" href="https://huggingface.co/fffiloni/spectrogram-to-music?duplicate=true"><img src="https://img.shields.io/badge/-Duplicate%20Space-blue?labelColor=white&style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAABAAAAAQCAYAAAAf8/9hAAAAAXNSR0IArs4c6QAAAP5JREFUOE+lk7FqAkEURY+ltunEgFXS2sZGIbXfEPdLlnxJyDdYB62sbbUKpLbVNhyYFzbrrA74YJlh9r079973psed0cvUD4A+4HoCjsA85X0Dfn/RBLBgBDxnQPfAEJgBY+A9gALA4tcbamSzS4xq4FOQAJgCDwV2CPKV8tZAJcAjMMkUe1vX+U+SMhfAJEHasQIWmXNN3abzDwHUrgcRGmYcgKe0bxrblHEB4E/pndMazNpSZGcsZdBlYJcEL9Afo75molJyM2FxmPgmgPqlWNLGfwZGG6UiyEvLzHYDmoPkDDiNm9JR9uboiONcBXrpY1qmgs21x1QwyZcpvxt9NS09PlsPAAAAAElFTkSuQmCC&logoWidth=14" alt="Duplicate Space"></a>       
            <a href="https://colab.research.google.com/drive/1FhH3HlN8Ps_Pr9OR6Qcfbfz7utDvICl0?usp=sharing" target="_blank"><img src="https://colab.research.google.com/assets/colab-badge.svg" /></a>
        </span>
    </p>
    
    <p style="font-size: 0.8em;line-height: 1.2em;border: 1px solid #374151;border-radius: 8px;padding: 20px;">
    About the model: Riffusion is a latent text-to-image diffusion model capable of generating spectrogram images given any text input. These spectrograms can be converted into audio clips.
    <br />—
    <br />The Riffusion model was created by fine-tuning the Stable-Diffusion-v1-5 checkpoint.
    <br />—
    <br />The model is intended for research purposes only. Possible research areas and tasks include 
    generation of artworks, audio, and use in creative processes, applications in educational or creative tools, research on generative models.

    </p>
    
    <div class="footer">
        <p>
        <a href="https://huggingface.co/riffusion/riffusion-model-v1" target="_blank">Riffusion model</a> by Seth Forsgren and Hayk Martiros - 
        Demo by 🤗 <a href="https://twitter.com/fffiloni" target="_blank">Sylvain Filoni</a>
        </p>
    </div>

    <div id="may-like-container" style="display: flex;justify-content: center;flex-direction: column;align-items: center;">
        <p style="font-size: 0.8em;margin-bottom: 4px;">You may also like: </p>
        <div id="may-like" style="display:flex; align-items:center; justify-content: center;height:20px;">
            <svg height="20" width="158" style="margin-left:4px">       
                <a href="https://huggingface.co/spaces/fffiloni/img-to-music" target="_blank">
                    <image href="https://img.shields.io/badge/🤗 Spaces-Image to Music-blue" src="https://img.shields.io/badge/🤗 Spaces-Image to Music-blue.png" height="20"/>
                </a>
    </svg>
        </div>
    </div>

"""

css = '''
    #col-container, #col-container-2 {max-width: 510px; margin-left: auto; margin-right: auto;}
    a {text-decoration-line: underline; font-weight: 600;}
    div#record_btn > .mt-6 {
        margin-top: 0!important;
    }
    div#record_btn > .mt-6 button {
        width: 100%;
        height: 40px;
    }
    .footer {
        margin-bottom: 45px;
        margin-top: 10px;
        text-align: center;
        border-bottom: 1px solid #e5e5e5;
    }
    .footer>p {
        font-size: .8rem;
        display: inline-block;
        padding: 0 10px;
        transform: translateY(10px);
        background: white;
    }
    .dark .footer {
        border-color: #303030;
    }
    .dark .footer>p {
        background: #0b0f19;
    }
    .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; width: 13rem;
    }
    #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: 0.25rem !important; padding-bottom: 0.25rem !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;
    }

'''
 


with gr.Blocks(css="style.css") as demo:
    
    with gr.Column(elem_id="col-container"):
        
        gr.HTML(title)
        
        prompt_input = gr.Textbox(placeholder="a cat diva singing in a New York jazz club", label="Musical prompt", elem_id="prompt-in")
        audio_input = gr.Audio(sources=["upload"], type="filepath", visible=False)
        with gr.Row():
            negative_prompt = gr.Textbox(label="Negative prompt")
            duration_input = gr.Slider(label="Duration in seconds", minimum=5, maximum=10, step=1, value=8, elem_id="duration-slider")
            
        send_btn = gr.Button(value="Get a new spectrogram ! ", elem_id="submit-btn")
            
    with gr.Column(elem_id="col-container-2"):
        
        spectrogram_output = gr.Image(label="spectrogram image result", elem_id="img-out")
        sound_output = gr.Audio(type='filepath', label="spectrogram sound", elem_id="music-out")
        
        #with gr.Group(elem_id="share-btn-container"):
        #    community_icon = gr.HTML(community_icon_html, visible=False)
        #    loading_icon = gr.HTML(loading_icon_html, visible=False)
        #    share_button = gr.Button("Share to community", elem_id="share-btn", visible=False)
        
        gr.HTML(article)
    
    send_btn.click(predict, inputs=[prompt_input, negative_prompt, audio_input, duration_input], outputs=[spectrogram_output, sound_output])
    #share_button.click(None, [], [], _js=share_js)

demo.queue(max_size=250).launch(debug=True)