V.2.0Had some pics lying around, sorting and deleting images at the moment to get more structure. Mostly added sculptures, mask's, stupid costumes etc.<hr />Finally found a trainer for Chroma Lora's that works for me, with a few minor modifications. Best part, i can finally see the progress while training without guesswork as it always was with Flux based Lora's. Didn't think it would work so well, can go higher with the DIM, Resolution etc. and still resources left on a 16GB card.Linky: <a target="_blank" rel="ugc" href="https://github.com/Akegarasu/lora-scripts">https://github.com/Akegarasu/lora-scripts</a>Just a little test. Mostly sculptures of monster figures and a few drawings. Bit bland because of steriel backgrounds and plasticy nature. Also, hit or miss since it would still need more training, but a lot closer to the original images then just normal Flux Lora's.<s>Haven't tested if it has the same behaviour on normal Flux checkpoints as basic Flux Loras have on Chroma, where Comfy just spits out a few missing key/block errors, but it still kinda works.</s>Well, seems to work on normal Flux models ( Krea ) without any errors, at least not in Comfy. Doesn't seem to have much impact though, but, that's maybe a good thing in some cases.Yeah, and rate everything that has the word "tentacle" in it XXX ... jesus christ.

Files with SHA256: 8d5fa5bbf97908eb902a7bc8afdcf532f701d6ac413a9982f9b4b6cd2e32022e