Sure - I'll always take some pointers. I've had a number of successes training for Flux, but where I find the model resistant to what I want to teach it is in painting styles. It gets weird when you try to teach it surface textures. Which is a large part of what I want a paintng model to capture.

So my oil painting model for Flux is pretty good, but after quite a few attempts with varying datasets and merges between different training, it never quite gets the surface texture of the SDXL model. Better rendering for sure, but too many places where I expect to see brush strokes, I get instead just a kind of ugly, grainy noise.

And this I find much more strongly with trying to teach Flux impressionism (specifically, the narrow focus of impressionism that most interests me, such as Monet's most abstract and impasto work. I.e., his Rouen cathedral studies and hay bales studies.)

Any other model where the focus is less on surface texture of paint, and instead is closer to digital art, I have no issues at all training Flux.