OneRing (aka GirlsAi)

All preparatory procedures are finished (dataset increased, prompts manually improved, modern improvements added, which I reached through practice + training will take place on the new rouwei, which should give some advantages, including understanding of natural language). Training as usual will take a long time, but in the end I am almost sure this will be the best product I have ever made.

Guys, I haven't disappeared, I just have serious problems with access to my computer. There was a fire in my apartment and a lot was damaged (in addition to this, there were family problems and the arrival of distant relatives - I couldn't even work at my main job). I'll be back soon, don't worry. Good people are helping me and everything will be fine. Everything that happens is for the best - I believe in it.

Hi guys, I have some news, I was able to squeeze in training WITHOUT COMPROMISE in 1536x1536, the native resolution of illustrious v1.1. This is a test lore with complex anatomy, the first step to mastering the highest quality at the moment, I have already trained on illustrious v1.1, but at standard resolution and noticed serious problems with color and details when increasing the resolution during image generation. I am not sure that this problem is related specifically to the resolution of the training process, but most likely yes. In the end, I decided to test the strength of my video card (and time) on a relatively small lora and if everything goes according to plan, then I will raise the quality bar even higher than before (however, so far only epsilon, not v-pred)

Training this lora of 221 images will take more than a day (such are the increased requirements these days), so wait for updates :3