alt Hacker NewsI've been running it through tests on my local machine. The Text-to-Image component is VERY poor. It's not a good sign when the stock ComfyUI Flux3 Klein template with built-in config/prompts produces SD 1.5 levels of half-melted faces and various other body horror.
Flux 2 Klein 9b also had anatomical issues as well, but this one feels practically lobotomized.
That looks worse than Flux 2.
What the heck are they doing?
Is this fast/cheap at the pareto frontier? Doesn't really work if humans look deformed.