← sources

Tips / settings for better results with MOSS-Image Base + Turbo LoRAs? (blur / out-of-focus / smoothness issues)

source: reddit (reddit:StableDiffusion)published 2026-11-03 10:16ingested 2026-11-04 02:39

retained text

Anyone have solid settings or tips for getting cleaner output from MOSS-Image Base + its turbo LoRAs?

Important clarification: I am strictly talking about the Base model, not SFT.

I’m not interested in SFT  it tends to give a more artificial / over-polished look, and I don’t want to stack 3 extra LoRAs just to have a raw/natural feel.

With Base + turbo LoRAs I’m frequently getting:

Blur in the wrong places

Soft / out-of-focus areas

Over-smoothness / loss of local detail

These issues are much milder (or easily fixed with negatives) when running plain Base(i usually run base at cfg 4 and 35-50 steps with sa\_solver/Euler+sgm uniform). Once the turbo adapters are loaded they become stubborn, which tracks with the low-CFG / CFG-1 setups most turbo recipes use.

I’ve already experimented with:

LoRA strengths 0.5 / 0.6 / 0.7 / 0.8 / 0.9

Higher CFG (2 / 2.5 / 3 / 3.5)

Higher steps (up to 16)

Still no reliable fix.

Looking for any settings, specific turbo LoRA (4-step, 8-step DMD, DMD2, etc.), sampler/scheduler/shift combos, or other workflow tweaks that actually keep sharpness and detail on Base.

Would genuinely love to hear if anyone has found a better recipe that works.

document fetched 2026-11-04 00:31 · canonical https://reddit.com/r/StableDiffusion/comments/1xmpadq/tips_settings_for_better_results_with_mossimage