Another small experiment.

This merge uses a few new LoRAs I trained, along with a deturbo base prepared by Ostris

This isn’t a turbo model. It’s slower, but rewards you with cleaner detail if you let it breathe.

Recommended settings:
• Euler or DPM++ 2s a RF
•Recommended: 30–40 steps (higher steps improve detail; 50+ for maximum fidelity).
Works at 25, but quality improves significantly with more steps.
• CFG ≈ 3
If you use Comyfy UI get this https://github.com/erosDiffusion/ComfyUI-EulerDiscreteScheduler seems great.

Have fun exploring.