Can you please upload a safetensor version of V3? :)

Absolutely stunning! Good work!

目前为止 个人认为 真人效果最好的

what model and vae did you use for them? thanks

I liked the faces of v1 more than v2 so i kept using the former, but v3 seems like a proper continuation and an upgrade.
My experience with v1 so far was that if I put only a single "ulzzang" in positive prompt and leave negative empty, it would most often give the face that i prefer. Adding literally anything in both negative and positive prompt would result in less images to my liking. Also hiresfix didn't seem to improve the image at all. Didn't work well with lower qualities loras.
V2 was better at handling longer prompts, but, again, i didn't like V2 faces. V3 seems to get as good faces as V1 did and can handle longer prompts. Can also use hiresfix with it; didn't try loras yet.
Two notable problems for me:
1. red color oversaturation, that was also present in V2. Way too red lips, way too red blush (especially with hiresfix, even makes black/brown hair turn red), clothes tend to default towards red (same was with v1 but towards blue).
2. inflatable balloons seem to get a bit out of hand. With every version breasts become bigger and bigger. While V1 could have a variety of sizes and more often than not actually generate smaller breasts, V3 seems to dedicate half of the pixels towards huge boobas. Just not my preference.

Thanks for your work.

效果贼拉好

Could you add a safetensors version,plz？

Please add a safetensor version.

用这个模型跑机甲 效果居然挺好😂

這是目前為止最棒的，期待V4版本~