📢 Update Notice

v2 version added:

This version builds upon v1 by introducing training data for suck2foot, aiming to support dual-foot interaction.

For clarity and ease of use, the original v1 LoRA has been re-uploaded with a new filename, If you don't need suck2foot, feel free to keep using v1 — its suck1foot performance is generally better than in v2.

📘 LoRA Overview

All training images use a 1216×832 resolution (horizontal).
Vertical or smaller resolutions aren’t recommended — unless you enjoy making things hard for yourself.
(Which, clearly, I do.)🤣

Compared to Part 1, this version removes many redundant tags.

Toes turned out to be more trouble than they’re worth, So for v1, I slapped a single_shoe or single_boot on everything. Not a weird kink, I promise — just trying to save myself some headaches, lol.

📌 Recommended Settings

• Known issue: Using tongue_out in your prompt can sometimes interfere with expression-related tags (they might behave unpredictably). Therefore, it’s recommended to lower the weight of tongue_out to 0.8 and raise the weights of other expression prompts to 1.3.

• Best used in combination with a character LoRA.

• Use DPM++ 2M Karras sampler and a resolution of 1216×832 for best results.

• Trigger word: suck1footor suck2foot.

• For optimal performance, pair with single_shoe or single_boot.

Check the sample images for more info.