Will this be an already "aesthetically tuned" version?
So Lode said Chroma.1 is not an aesthetically tuned model, intentionally leaving that for fine-tuners. Is it the same for this version?
The trouble is that different people have wildly different ideas about what is and isn't aesthetic. So many descriptions of AI theories brag about their aesthetics ranking and yet there is a central fallacy to this: Absolutely no one and certainly no machine can be the universal arbiter of art. Individuality is real. That means individuals have individual taste in aesthetics.
That said, some things, like third limbs and random body parts that shouldn't exist, are no-brainers when it comes to aesthetic tuning.
aesthetic tune is a "biased dataset"
and no this is not tuned to be aesthetic either, this is purely "re pretrain" to introduce the missing distribution in the model while preserving the existing distribution and widening the variance so people can "collapse" the model into more aesthetically pleasing one using specific lora or RL ontop easily.
Any plans on making an e621/danbooru/ect. finetune yourself? 🤔