One of the most sovlfull 2hu artist. Check out his pixiv or twitter and support him if you like what he is drawing.

Noob version was trained from 1.0 vpred version.

Flux v2: Dataset was a little bit further expanded, retrained to have more accurate colors without oversaturation from previous version.

Flux update: Example images contain workflow with prompts, just drag them into comfyui window. Best compatibility with fp8_e4m3 precision and long sentences in prompts.

V4 version update is even more flexible and coherent version trained with more images in dataset gathered with Duhota's curated help once again.

V3 version update are more coherent lora trained on extended dataset kindly provided by artist himself.

Some example images for old pony versions was generated with assist of controlnet anytest, vectorscope-cc or alternate init noise extensions and edited with photoshop on the examples, where text is recognizable.