[FEEDBACK] Enhance cluster playlist naming promptΒ #238
Description
Hi! I'm loving AudioMuse-AI so far, the integration with Jellyfin and Finamp is great!
I have a few suggestions to modify the cluster playlist naming prompt. My goal is to nudge the model towards generating more consistent playlist names.
Original prompt: (Version v0.8.2)
You're an expert of music and you need to give a title to this playlist.
The title need to represent the mood and the activity of when you listening the playlist.
The title MUST use ONLY standard ASCII (a-z, A-Z, 0-9, spaces, and - & ' ! . , ? ( ) [ ]).
No special fonts or emojis.
* BAD EXAMPLES: 'Ambient Electronic Space - Electric Soundscapes - Emotional Waves' (Too long/descriptive)
* BAD EXAMPLES: 'Blues Rock Fast Tracks' (Too direct/literal, not evocative enough)
* BAD EXAMPLES: 'π―ππ πππππ π
πππ' (Non-standard characters)
CRITICAL: Your response MUST be ONLY the single playlist name. No explanations, no 'Playlist Name:', no numbering, no extra text or formatting whatsoever.
This is the playlist: - <Title of song 1> by <Artist name>
- <Title of song 2> by <Artist name>
- <Title of song 3> by <Artist name>
- <Title of song N> by <Artist name>
Suggested changes
- Line 1: Refine the role of the model and include MUST keyword -
You are an expert music collector and MUST give a title to this playlist. - Line 2: Include MUST keyword, and tweak words around "listen" -
The title MUST represent the mood and the activity of when you are listening to the playlist. - Line 4: Include the hardcoded character limits in the prompt -
The title MUST be within the range of 5 to 40 characters long. - Line 10: add a linebreak between the end of the prompt and the list of songs
Modified prompt:
You are an expert music collector and MUST give a title to this playlist.
The title MUST represent the mood and the activity of when you are listening to the playlist.
The title MUST use ONLY standard ASCII (a-z, A-Z, 0-9, spaces, and - & ' ! . , ? ( ) [ ]).
The title MUST be within the range of 5 to 40 characters long.
No special fonts or emojis.
* BAD EXAMPLES: 'Ambient Electronic Space - Electric Soundscapes - Emotional Waves' (Too long/descriptive)
* BAD EXAMPLES: 'Blues Rock Fast Tracks' (Too direct/literal, not evocative enough)
* BAD EXAMPLES: 'π―ππ πππππ π
πππ' (Non-standard characters)
CRITICAL: Your response MUST be ONLY the single playlist name. No explanations, no 'Playlist Name:', no numbering, no extra text or formatting whatsoever.
This is the playlist:
- <Title of song 1> by <Artist name>
- <Title of song 2> by <Artist name>
- <Title of song 3> by <Artist name>
- <Title of song N> by <Artist name>
Additional notes
Task Parameters:
All analysis and clustering parameters were default from version v0.8.2, with the exception of using ollama with granite4:7b-a1b-h
What was confusing or could be improved?
- There is no indication in the Web UI that clustered playlists with fewer than 20 songs are discarded. Given the default number of clusters (40 - 100) and a libary size of 700 songs, every cluster had less than 20 songs and was discarded. A notification to the user of 0 successful playlists and a suggestion to lower the K-Means min/max clusters would reduce user confusion.
- The hamburger menu icon to open the sidebar menu is small, I did not notice it for a long time.
- A dark mode would help tired eyes at night.
Any Other Thoughts or Ideas?
AudioMuse-AI is a great project, and i'm looking foward to it's continued success!