Library
Every adapter you have trained, with what it is for and how to use it.
Each card names the base model it was trained against, its size, how many steps it ran for, and — for a subject adapter — the trigger phrase to put in your prompt. Filter by Subjects or Styles, search by name or tag, and tag them yourself.
Training saves snapshots along the way, and they all appear here:
_lora-step00000200, -step00000400, and so on.
That is deliberate. More steps is not automatically better. An adapter usually gets sharper for a while and then starts to overcook — it stops looking like your subject and starts looking like your training photos, including their backgrounds and lighting. The only reliable way to find the good one is to render the same prompt with two or three and look.
Use in a render arms an adapter on the Generate screen.
Adapters land in your models folder, and also in
datasets/<name>/output/ where a training run writes them. The Library reads
both, so a run you launched yourself with kohya sd-scripts shows up here
alongside anything started from the app.