Type | |
Stats | 213 |
Reviews | (14) |
Published | Feb 5, 2024 |
Base Model | |
Training | Steps: 8,312 Epochs: 8 |
Usage Tips | Clip Skip: 1 Strength: 0.8 |
Trigger Words | l3ra |
Hash | AutoV2 B8640ECDBE |
Trained on 150 publically available photos from @lerabuns and her other accounts at instagram.
This is my first model, so it's not perfect.
It was trained with Kohya on the AbsoluteReality Checkpoint (v1.81) using the AdamW optimizer, and Clip Skip 1.
The activating tag is "l3ra". In the training images, it was always used in conjunction with "1girl" and "solo".
The results are better when adding more descriptive tags.
For example, as for hair, the tags "blonde hair", "brown hair" and "multicolored hair" have been used.
You may sometimes have to slightly tag for "blue eyes" in order to get the correct eye color.
As for the weight, anything from 0.6 to 1.0 seemed to work fine. Starting with a 0.8 seems to give decent results.
This model also works to some degree with Anime Checkpoints. Use it with a low weight (like 0.6) in that case.