Moss Image FinalCut (12 - 25Step)

I strongly advise against using INT8 of this model

• Use at 12-25 steps sgm_uniform/er_sde (Beyond 25 quality tends to taper off)

• Up to 2048x2048

• Trained with both 4096 caption length and short WD14 tags - English language reinforcement and Pony style training.

Onsite generations look drastically worse then local comfy output

Comfy Quants

INT8 was changed to simple convrot which reduced canvas but is still far underperforming Universal Formats. This should not be the case but I have not seen any published info regarding this issue.

Compare the two Civitai Flashing breast images and decide for yourself, for me the slight speed increase is not worth the drastic quality loss