# Fizgig meets Music: Train and Tweak Ace-Step 1.5 Loras — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/82641d1af2f400a8

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-11 23:23 ingested 2026-10-12 02:38

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1x3mku9/fizgig_meets_music_train_and_tweak_acestep_15/)

## retained text



```
Fizgig, my free open-source LoRA trainer, now does music.

Train ACE-Step 1.5 XL LoRAs on your own songs and play every checkpoint in the gallery as it trains.  
You can also make slider LoRAs that dial between two styles, like calm to energetic, without needing any songs.

Repair Studio works like a DAW for AI music, with per-block sliders, covers, repaint and extend, and LoRA mixing, **and saving out wavs of the music you create in it.**

A new I'm feeling lucky button connected to the Repair Studio which works ludicrously well with the DAW-like mode for Ace-Step for exploration purposes.

To get started there's a free and fully credited chiptune dataset and LoRA - **I strongly encourage** reading the tips I've included in it - starting training this way will make everything super clear: [https://huggingface.co/datasets/ShootTheSound/Fizgig-ACE-Step-Chiptune-Starter](https://huggingface.co/datasets/ShootTheSound/Fizgig-ACE-Step-Chiptune-Starter)

Release notes: [https://github.com/shootthesound/Fizgig/releases/tag/v7.3.0](https://github.com/shootthesound/Fizgig/releases/tag/v7.3.0)
```



document fetched 2026-10-12 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1x3mku9/fizgig_meets_music_train_and_tweak_acestep_15
