Models trained with more parameters have more knowledge and capabilities than models trained with less parameters. Right now Flux is the publicly available with the most parameters, but the volume makes training prohibitively expensive and resource intensive. Unfortunately without monetization it's hard for anyone to justify the cost of training.

>>8662490
It's only at v0.41 current though. I presume the intention is to make 1.0 the initial release build so it seems like there's more than halfway to go for that yet.

>>8662493
>Also add the ridiculous gen times on anything that isn't 5090 or better.

I've only got a 3090 but I've run Flux and though it takes longer than SDXL it's not that bad. Maybe a minute or couple per gen? It's been a while since I've tried since that was with the base Flux model which is censored but it was impressive to see the prompt adherence and how it could generate legible text and such.