Suno Custom Models: build a model from your own songs
Published September 19, 2026. Official documentation checked on the same date.
Use a Custom Model when you have a collection of your own music and want to explore a consistent musical direction. The useful preparation is choosing what those tracks have in common, so you can later judge whether the result reflects that direction.
Requirements and setup
Suno's September 9 guide says Pro and Premier users can create up to three private models, starting with at least six songs they own the rights to. In Create, open the model picker, choose Create Custom Model, add the tracks and submit. When processing finishes, select the new model from the picker. Models cannot be shared between users. Source: Using Custom Models.
The current v6 documentation includes Custom Models among its supported creation tools. Older tutorials describing a v5.5-only workflow should be read alongside the current model picker. Source: What's new in v6?.
Choose a musical direction for the reference set
The following selection exercise is our suggestion, not a published Suno training formula. Listen to your candidate tracks and complete this sentence: “These belong together because they share …”
Useful answers are audible: dry intimate vocals, compact arrangements, rounded bass, or a particular rhythmic feel. “They are all my songs” identifies the owner but does not help you compare the musical result.
| Reference-set idea | What to evaluate later |
|---|---|
| Sparse acoustic songs | Whether a new arrangement leaves similar space. |
| Rhythm-focused instrumentals | Whether the groove fits your intended direction. |
| Dense electronic productions | Whether the texture supports the song instead of covering it. |
Write down the tracks in each set. If you later replace several references, give that experiment its own note so you can distinguish it from a prompt change.
Compare a Custom Model against a baseline
Choose one short song brief and keep the lyrics, reference audio and other visible settings stable. Generate with a standard model, then repeat with your Custom Model. Record the actual names of both models.
This original brief is deliberately narrow:
Mid-tempo alternative R&B, rounded bass, restrained drums,
close lead vocal, an uncluttered verse and a wider chorus.
It is an untested writing example. Define what success means before listening: perhaps more consistent low-end character or a verse that better matches the space in your reference tracks. Do not reduce the comparison to “the personalized one must be better.”
If one result has a useful texture and an unsuitable melody, record those separately. A model can help one aspect of a brief without becoming your default for every song.
Custom Models, Voices and My Taste
These features give you different things to evaluate. Use Voices for a voice-profile workflow. Use the Custom Model experiment above when comparing a set of musical references.
My Taste personalizes the Style text generated by the Magic Wand using your preferences. If you use it during a model comparison, save the resulting text and keep it fixed across the two runs. Source: My Taste.
Do not treat a model change and a rewritten Style description as one controlled experiment. You need to know which part changed before deciding what helped.
Common questions
Do I need one Custom Model for every genre?
Start with one reference set and one clear use case. Build another experiment only when the first set cannot answer the next musical question. A genre label alone does not tell you whether the references are coherent.
Will a Custom Model reproduce one of my songs?
Do not use this workflow as an exact recreation promise. Keep the original audio when a specific performance matters. For a local correction to that performance, use Replace Section.
Prepare your briefs and comparison notes in MusicPrompt Studio. Creating, storing and selecting the Custom Model happens in Suno; MusicPrompt records your text decisions and listening results.