AuraFlow Reviews

Seems like a very promising start for this model... it is heavier than sdxl and still lacks the style and refinement of the best that that base has to offer, but prompt adherence is noticeably better, actually being able to add some short text is nice and I am guessing the image quality and refinement/upscale issues can be fixed with time.
My one concern is with the lazy lobotomy job for nuking its understanding of.. certain anatomy. If it had to be done for legal reasons, fine, nothing to be done about the law, but if it was for other reasons please reconsider as I'd argue it does little to "safeguard" things (since anyone can just inpaint train a lora or merge to add that stuff in, and even this base merge easily produced stuff that would get me banned here on civit when I put it through my limit/safety test prompt rubric) and makes the model more dangerous (since it does not know these things I had it throw some bits and pieces into images randomly (not explicit anatomy, but the logical continuation from it and things like that), and I often could not get rid of it with negatives since the model seems to have had no idea what it is) while also reducing model overall quality since at least from my experience there is always collateral damage when trying to remove things from a model. For SD1.5 it might have been ok to just eliminate the knowledge, but for a model this capable of following prompts, I would much rather just be able to tell it what I do not want in the negatives instead of it having holes in its digital brain...