Trained similarly to full model for floral patterns: https://civitai.com/models/26295/floral-patterns

Join our Discord for any questions or feedback!

Training process

First attempt at training a LoRA, so any feedback would be appreciated. LoRA was trained using Kohya’s Dreambooth Lora Colab on top of AnyLora and an anime VAE. 

LoRA is best used with weight ~0.8 - weights that are too high can lead to an overcooked look. Due to size, it seems the Lora is less able to handle complex details. Images generated using it will generally look more simplistic than the full model.

Prompting

Use the word "pattern" to trigger the style. You can also use negative prompt to get rid of unwanted colors if the positive prompt isn't providing enough control.

For best results, use the following pattern - a [COLOR1] pattern with [SUBJECT] on a [COLOR2] background

Example prompts

• A pattern with lilacs and birds on a pink background

• A black and white pattern with butterflies

• A blue and yellow pattern with flowers

Settings

Resolution must be 768 x 768 for best results

Sample images were generated with CFG = 11 and 75 inference steps. 

Inference settings otherwise shouldn’t matter too much.