app.py

import torch
import torchaudio
import gradio as gr
import tempfile
import time

from stable_audio_tools import get_pretrained_model
from stable_audio_tools.inference.generation import generate_diffusion_cond

# Load model
device = "cuda" if torch.cuda.is_available() else "cpu"
model, model_config = get_pretrained_model("RoyalCities/Foundation-1")
model = model.to(device)
sample_rate = model_config.get("sample_rate", 44100)

def generate_audio(prompt, negative_prompt="low quality, noise, distortion, muffled",
                   duration=15, steps=100, guidance_scale=7.0, seed=-1):
    start = time.time()
    duration = max(1, min(47, float(duration)))
    steps = max(20, min(200, int(steps)))

    conditioning = [{"prompt": prompt, "seconds_start": 0, "seconds_total": duration}]

    with torch.inference_mode():
        output = generate_diffusion_cond(
            model,
            conditioning=conditioning,
            steps=steps,
            cfg_scale=guidance_scale,
            sample_size=sample_rate * int(duration),
            device=device,
        )

    audio = output.squeeze(0).cpu()
    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    torchaudio.save(tmp.name, audio, sample_rate)

    elapsed = time.time() - start
    return tmp.name, f"Generated {duration:.0f}s in {elapsed:.1f}s | Steps: {steps}"

with gr.Blocks(title="Foundation-1 × SPARK M!NDS") as app:
    gr.Markdown("# Foundation-1 × SPARK M!NDS")
    with gr.Row():
        with gr.Column(scale=2):
            p = gr.Textbox(label="Prompt", placeholder="dark trap beat, 808s, 140 bpm", lines=3)
            np_ = gr.Textbox(label="Negative", value="low quality, noise, distortion", lines=2)
            with gr.Row():
                d = gr.Slider(1, 47, value=15, step=1, label="Duration (sec)")
                s = gr.Slider(20, 200, value=100, step=10, label="Steps")
            with gr.Row():
                g = gr.Slider(1, 15, value=7.0, step=0.5, label="Guidance")
                sd = gr.Number(value=-1, label="Seed (-1 = random)")
            btn = gr.Button("Generate", variant="primary", size="lg")
        with gr.Column(scale=1):
            ao = gr.Audio(label="Output", type="filepath")
            io = gr.Textbox(label="Info", interactive=False)
    btn.click(fn=generate_audio, inputs=[p, np_, d, s, g, sd], outputs=[ao, io])

app.launch()