assuplyingdown

This is my first LORA, trained off only 11 images, so it's probably rubbish, but it seems to do what I intended, so that's a start.

I was looking to reproduce a specific pose, where the subject has her ass in the air, and her knees under, but more lying down than crawling/all fours/etc. Since it is only 11 images, using it at a lower weight yields more diverse generations, but seems to stick to the pose requested. The activation text is auok AssUpOnKnees

It also has a (mostly unintended) tendency towards tattoos and a hand under the chin. We don't make mistakes, we create learning opportunities!

Updated version v0.2:

After further testing with the CivitAI generated LORA, I noticed some things and decided to try training on my local SD. I added another 14 hopefully helpful images to the dataset and trained the Lora according to this helpful post: https://i.imgur.com/USDCHQF.png

The new version seems to handle perspectives and framing a bit better, but it still mangles feet. Anybody got a clue on what that might be?

Updated version PDXL:

Ran the existing dataset through the Civitai LoRA training blender set for "Pony", and found it worked decently.