Seamlessly insert an advertisement into any song using MusicGPT AI.

• Prompt enhancement – Type a short product description (e.g. "organic coffee subscription") and click Enhance ✦ to let Gemini rewrite it into a rich, MusicGPT-friendly prompt (e.g. "A punchy 15-second upbeat advertisement for an organic coffee subscription, warm acoustic pop with a friendly female vocal, evokes morning freshness and artisanal quality.").
• Seamless mode – MusicGPT's Inpaint endpoint replaces a chosen time window in the song with AI-generated ad content that musically blends with the surrounding audio.
• Local splice mode – Classic pydub crossfade/volume-ducking splice. Supports both uploaded ad audio and AI-generated ads from a text prompt.

1. Sign up at musicgpt.com — new accounts get $20 free credits.
2. Go to your dashboard / API settings and copy the key.

1. Visit aistudio.google.com and sign in with your Google account.
2. Create an API key and copy it.

The Gemini key is only required for the prompt enhancement feature. If you leave it blank the Enhance button will return an error, but all other functionality still works.

# macOS
brew install ffmpeg

# Ubuntu / Debian
sudo apt-get install ffmpeg

pip install -r requirements.txt

cp .env.example .env
# then edit .env and paste your keys:
#   MUSICGPT_API_KEY=<your_musicgpt_key>
#   GEMINI_API_KEY=<your_gemini_key>

uvicorn app.main:app --reload

Open http://127.0.0.1:8000 in your browser to use the web UI.

Interactive docs: http://127.0.0.1:8000/docs

{
  "raw_prompt": "organic coffee subscription",
  "ad_length_seconds": 15
}

Response:

{
  "enhanced_prompt": "A punchy 15-second upbeat advertisement for an organic coffee subscription, warm acoustic pop with a friendly female vocal, evokes morning freshness and artisanal quality."
}

{
  "song_id": "<file_id from /upload/song>",
  "ad_mode": "audio_upload",
  "ad_id": "<file_id from /upload/ad>",
  "insert_at_seconds": 30,
  "seamless_integration": true,
  "replace_window_seconds": 15,
  "ad_integration_prompt": "A punchy 15-second coffee-app ad, upbeat pop",
  "gender": "female"
}

Ad modes

Key fields

# Seamless integration with a text-prompt ad:
python testbench.py song.mp3 \
    --text-prompt "A catchy 15-second upbeat jingle for a coffee delivery app" \
    --insert-at 30 --window 15

# Seamless integration with uploaded ad audio:
python testbench.py song.mp3 --ad ad.mp3 \
    --insert-at 45 --window 12 \
    --prompt "A punchy advertisement blending with the surrounding pop music"

# Local splice with uploaded ad:
python testbench.py song.mp3 --ad ad.mp3 --insert-at 60 --no-seamless

# Local splice with AI-generated ad:
python testbench.py song.mp3 \
    --text-prompt "An energetic sneaker brand ad" \
    --insert-at 30 --no-seamless --music-style Electronic

Run python testbench.py --help for all options.

User types product description (e.g. "organic coffee subscription")
        │
        ▼ [optional] POST /enhance-prompt
Gemini rewrites it into a rich MusicGPT-friendly prompt
        │
        ▼
User uploads song + (enhanced prompt or ad file)
        │
        ▼
seamless_integration?
    ├── YES → POST /v1/inpaint (MusicGPT)
    │          audio_file = song
    │          replace_start_at / replace_end_at
    │          prompt = enhanced ad description
    │          → poll GET /v1/byId → download final MP3
    │
    └── NO  → ad_mode?
              ├── text_prompt → POST /v1/MusicAI → poll → download ad
              │                 then pydub crossfade + duck
              └── audio_upload → pydub crossfade + duck directly