lora-trainer

On the rail as lora-trainer.

When to use it

Teaching the tools a look or a subject of your own from a small set of images, so it can be called up by name in a prompt afterwards.

When to reach for something else

It takes no written direction and only learns from images you feed it, so it makes nothing by itself. It needs several consistent images, and the look only appears once you run it elsewhere.

How to use it

1. Write what you want in Name.
2. Set Base Model, Training Images on the card.
3. Press Run.

On the card

Control	What it does	Choices
Base Model	Which engine to build your style on. FLUX.1 trains on 4-20 images; FLUX.2 on 9-50.	FLUX.1 · fast, FLUX.2 · higher quality
Name	What this LoRA is called in the picker later. Yours only, so name it however you will recognise it.	Your words
Trigger Word	A short made-up word you drop into a prompt to call this LoRA up. Five to eight characters and not a real word, so nothing else can claim it.	Your words
Training Images	The example images to learn the style or subject from.
Run	Starts the run. The number on it is what this run costs in credits, worked out before you press.

Settings

These sit behind the settings button on the card.

Setting	What it does	Choices
Style Mode	On teaches a look or style; off teaches a specific subject or character.	On or off
Training Steps	How long to train. More can capture the style better but takes longer and can overdo it.	500 to 2000
Learning Rate	How fast it learns each step. The default is tuned; only change it if you know why.	0.00001 to 0.0005