Verified: 2 years ago
SafeTensor
Type | |
Stats | 321 404 |
Reviews | (51) |
Published | Apr 20, 2023 |
Base Model | |
Training | Steps: 6,000 |
Trigger Words | <lora:thirdlife2:1> 3DCG polygon rendering |
Hash | AutoV2 666202F362 |
This is a Lora that tries to mimic Second life in-game screenshots. It has mostly been trained on anime style avatars, the environment and objects. It works best at 0.8~1. For best results use UNIPC sampler and add "3DCG, polygon rendering" in the positive prompt and "2d drawing" in the negative prompt. "Pretty face" or "pretty eyes" in the prompt also seem to give better face results with authentic in-game look. Perhaps there are other prompts I am not aware of to make it seem less like a CGI prerendering or drawing. Doing a hires fix with latent upscale seems to fix eyes and disfigured body.
I was able to make some furries and human avatars as well (the word 'cat' seems to be the triggerword for furries, and mature male/female for humans)