Search Results
6/20/2025, 11:11:16 AM
>>105649588
I don't think that the problem is Chroma's dataset. It could just as easily be that Flux is a huge model, so it's hard to teach it every concept. Turns out that for realism, the model knows many, many things. But that comes at a cost. It's very difficult to teach it new concepts because of how massive its knowledge is. That's just how it is. I am personally comfortable with Chroma not being the top anime model. Two separate models for anime/realism aren't bad. I just don't like SDXL being used to train anime, and Lumina will solve that for now. We have yet to see Flux's full finetune potential, it's just that with the hardware lodestone has it would take a while to actually see insane results.
I don't think that the problem is Chroma's dataset. It could just as easily be that Flux is a huge model, so it's hard to teach it every concept. Turns out that for realism, the model knows many, many things. But that comes at a cost. It's very difficult to teach it new concepts because of how massive its knowledge is. That's just how it is. I am personally comfortable with Chroma not being the top anime model. Two separate models for anime/realism aren't bad. I just don't like SDXL being used to train anime, and Lumina will solve that for now. We have yet to see Flux's full finetune potential, it's just that with the hardware lodestone has it would take a while to actually see insane results.
Page 1