FLUX.2-klein-9B raw model bench — prompt goes straight to the model

Adjustable size and steps — prompt goes in verbatim. One render at a time.

Sketch bench: your words, untouched, into FLUX.2-klein-9B (diffusion Q4_0 + Qwen3-8B + flux2 VAE via sd.cpp). Pick size and steps per render; server clamps to 256–1024px and 1–8 steps. 9B on CPU is slow — expect many minutes per render. Only 80/443 public.