app.py

import os
import io
import requests
import numpy as np
import gradio as gr
from PIL import Image
from loguru import logger

riffusion_url = os.environ["MODEL_ENDPOINT"]

def bta(audio_content: bytes):
    return (44100, np.frombuffer(audio_content, dtype=np.int16))

def bti(image_content: bytes):
    return Image.open(io.BytesIO(image_content))


def riffusion(query):
    res = requests.get(url=riffusion_url, params={'query': query}).json()
    audio_url = res['audio_url']
    image_url = res['image_url']
    image_content = []
    audio_content = []
    logger.info(f'request query = {query} \naudio url = {audio_url} \nimage url = {image_url}')
    for a in audio_url:
        audio_content.append(requests.get(a).content)
    for i in image_url:
        image_content.append(requests.get(i).content)
    return bti(image_content[0]), bta(audio_content[0]), bti(image_content[1]), bta(audio_content[1]), bti(image_content[2]), bta(audio_content[2])

demo = gr.Interface(fn=riffusion, inputs="text", outputs=[
    "image", "audio", "image", "audio", "image", "audio"
    ])

if __name__ == "__main__":
    demo.launch()