Qwen 2512 is so Underrated, prompt understanding is really great, only Flux 2 Dev is better. I'm using Q4KS with 4-6 steps and it is fast (20-30 sec per gen), almost as fast as Anima model. It just need that LoRA love from the community.