> <

I made this LoRA because not even anime checkpoints seem to be able to generate this expression on their own.

Usage (important!)

Compatibility

Should work with most anime-like checkpoints. Tested with AOM3 (works best), and MeinaHentai. more testing will be done later.

Weight

I recommend 0.75~1.0; however, should you use other LoRAs with this, try a higher value such as 0.85~1.1

When in doubt, increase weight~! This LoRA impacts artsyle minimally.

Trigger Words

> < is the trigger word. Recommended to put emphasis on this, like (> <:1.2)

Also try one eye closed in your prompt with the above! (may be hard to generate from.)

Limitations

On lower weights and with other LoRAs, this may fail!

Avoid using prompts about the eyes, unless you're using one eye closed with this.

Bias to open mouth. Try (open mouth:1.2) in negative prompt if you need to get around this.

描述:

This is the initial release~

Training details
Images: Cherrypicked from gelbooru, with search tags >_< 1girl -1boy -bad_anatomy -bad_feet -bad_hands -animated; chibis are picked less to avoid bias of the model, and tags are not changed from original posts. Includes some R-18 content in the dataset.

Base Checkpoint: [925997e9] (animefull-final-pruned AKA NAI)

Parameters: Plug it into sd-webui-additional-networks extension and check.

训练词语: (> <:1.2)