AI Models We Use
Last updated: 2026-10-08
Melunox is an independent product built on third-party AI models. This page lists the models behind each feature, so you know what made your work. Melunox is not affiliated with or endorsed by these providers; their names and trademarks belong to them. When a model is unavailable, a request may be handled by the other model listed for the same feature.
1. Songs, vocals and instrumentals
- Suno V6 (Suno) — the main model for songs with vocals and for instrumentals.
- ACE-Step (open-source model by ACE Studio and StepFun) — used when Suno is unavailable.
2. Covers and reworks
- ACE-Step Audio-to-Audio (ACE Studio and StepFun) — re-performs a track in a new style.
3. Lyrics, style and music-video planning
- GPT-6 (OpenAI) — writes and edits lyrics, and drafts music-video storyboards.
- Qwen3.8 Flash (Alibaba Cloud) — turns a style you describe in any language into music tags.
4. Cover art
- GPT Image 2 (OpenAI).
5. Music videos and lip sync
- Wan 2.7 (Alibaba Cloud) — music-video clips, from text or from the previous frame.
- Sync Lipsync (Sync Labs) — matches mouth movement in a video to a song.
6. Stems, MIDI and sheet music
- Demucs HTDemucs (open-source model from Meta AI Research) — separates vocals, drums, bass and other stems, and makes instrumentals.
- Basic Pitch (open-source model from Spotify) — converts audio to MIDI.
- Audiveris and homr (open-source optical music recognition) — read sheet music from photos and PDFs.
- Slowed + reverb is ordinary audio processing (FFmpeg), not an AI model.
7. Content safety and labelling
The same safety checks apply whichever model handles a request: prompts and lyrics are screened before generation (our blocklist and Waffo Prompt Sift), and generated images and video frames are checked before they are shown. See the Acceptable Use Policy (melunox.com/legal/acceptable-use).
Everything these models produce is AI-generated, and public song pages are labelled as such.