This model is "Pixart Sigma 1024" version realistic model. (Not the 900M model.)Finetuning is not completely finished, but I think it will take more time, so I will share the beta version first. It gonna be update in few weeks in my ko-fi page.<a target="_blank" rel="ugc" href="https://ko-fi.com/dalcefo_artworks">Buy Dalcefo a Coffee. </a><a target="_blank" rel="ugc" href="http://ko-fi.com/dalcefo_artworks">ko-fi.com/dalcefo_artworks</a><a target="_blank" rel="ugc" href="https://ko-fi.com/dalcefo_artworks"> - Ko-fi ❤️ Where creators get support from fans through donations, memberships, shop sales and more! The original 'Buy Me a Coffee' Page.</a>I think Pixart Sigma has several problems. The most serious part of this was that the initial training data for the model itself was too small.And another problem is that the model is structurally too small. In conclusion,Pixart Sigma has really good facial and upper body expressions.However, the hands, feet and lower body is just a mess!i wonder it cloud overcome through community trainning? it seems very hard.This model's hands, feet and body expressions are also a mess.But it is much better than the original model and I would like to improve it further in version 1.0.I understand it is currently available on ComfyUI and SDNext. I'm not sure because I only use ComfyUI.Therefore, you must perform facedetail using SD1.5 or SDXL checkpoint depending on your system.T5xxl is required to operate the Pixart Sigma.You can download finetuned T5xxl-fp16 and fp8 above tap of this page.

Dalcefo_Pixart.Sigma-Lucida

dalcefoPixartSigma_t5xxlFp8ForLucida.safetensors

T5xxl-fp8 for Lucida

Files with SHA256: a257a4a4cbe9cdddae6b9db430c6f4bd4543f1aa816a1ba731300dc09ff37497