PLEASE READ:

This LORA was trained entirely from in-game screenshots as a "test run" to assess the efficacy of a segmentation workflow I set up for myself in ComfyUI. As such, this model lacks some of the flexibility that one might be accustom to from other models trained on a mix of sources. It has a very difficult time producing more cartoon-y or stylized results, and can look quite messy with models tuned for more "realistic" or rendered outputs.

Despite being trained entirely on a feral dataset, since, y'know, anthro odogaron isn't canon, it does still manage to produce anthro designs of both genders surprisingly well. I have noticed that it has a difficult time with perspective and poses for anthros though, with a strong preference for "all-fours" poses, for obvious reasons.

It also has a hard time getting the tail right, and this is because the segmentation model cut it off from a lot of images since the calcified regions of Odogaron's tail look very visually similar to rocks, meaning it mistook it for scenery and discounted it.

This model has ZERO Ebony Odogaron in the dataset. Zip. Zilch. None. If there's an interest in me doing so, I could make another version of this LORA specifically for Ebony, but that fell outside the scope of my testing here.

Trigger words: 0d0g4r0n, feral, monster, quadruped, scales, red body, red scales, fangs, claws, spines, tail

Replace "feral" and "quadruped" with "anthro" for an anthro look, the model usually should handle the rest