In-development models for bigASP 3
bigASP 3 is currently undergoing training. I'm dumping some of the intermediate checkpoints here for now. Load as a Flux 2 Klein 9B model. They generally aren't usable as-is. Still needs to undergo post training RL and I need to finish up the prompt enhancer model.
2026.08.17
Second attempt at RL is done, e2nvn3l1_checkpoint_00000090_v_old_comfyui.safetensors is the resulting LoRA which can be applied on top of 3fph2juh. Run with 20 step Euler or DPM2, no CFG needed (set guidance to 1 to disable cfg).
This makes the model 4x faster than base, and ups the quality and consistency of the model significantly.
Still requires detailed prompts for good results.
This is just the second attempt, so there's more for me to tweak.
2026.06.18
Main training is done, 3fph2juh_flux2klein9b is the last checkpoint from that phase. Next up: RL and prompt enhancer
