Skip to content

Improve playlist-title prompt variety#317

Open
markjohnson303 wants to merge 1 commit intoNeptuneHub:mainfrom
markjohnson303:patch-1
Open

Improve playlist-title prompt variety#317
markjohnson303 wants to merge 1 commit intoNeptuneHub:mainfrom
markjohnson303:patch-1

Conversation

@markjohnson303
Copy link

This PR updates the playlist title prompt used by AudioMuse. The previous prompt often produced repetitive titles that tended towards similar words and patterns (for example “neon”, “reverie”, “drive”).

The new prompt keeps the existing constraints (ASCII-only, single output, length bounds), but adds clearer guidance around originality and vibe, plus explicit “avoid trite phrasing” examples. In practice, this results in noticeably more varied, fun, and distinctive playlist names.

This prompt was tested with OpenAI gpt-5.2.

This PR updates the playlist title prompt used by AudioMuse. The previous prompt often produced repetitive titles that tended towards similar words and patterns (for example “neon”, “reverie”, “drive”).

The new prompt keeps the existing constraints (ASCII-only, single output, length bounds), but adds clearer guidance around originality and vibe, plus explicit “avoid trite phrasing” examples. In practice, this results in noticeably more varied, fun, and distinctive playlist names.

This prompt was tested with OpenAI gpt-5.2.
@rendyhd
Copy link
Contributor

rendyhd commented Feb 6, 2026

I've ran a test across multiple models and while it performs great on GPT 5.2, it fails on many others - especially on smaller non-thinking ones often used in self-hosting. I've attached the results.

I agree with you, the names it gives with your prompt are a lot better, I really like much better.

To make it work I'm thinking of a couple of steps:

  • Fine-tune the prompt to work on all thinking models.
  • Add a toggle in settings, where the user can switch between the thinking and non-thinking prompt (and maybe even add a customize button). This is because a prompt aimed for a thinking model will remain problematic for non-thinking models, they generally handle contradicting instructions (pick 5, pick 1), negative instructions (do NOT), and longer prompts quite bad. Because of the model limitation I think a toggle is better than over complicating a system prompt and making it longer.

ai_naming_OG_Prompt.html
ai_naming_PR_Prompt.html

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants