9 LoRAs vs. 50-step baseline - Comparison of different Moss SFT turbo LoRAs
retained text
I tested every MOSS-Image turbo LoRA I could find against the plain 50-step model. I wanted to know whether distilled models can still render the surface of classical paintings: craquelure, canvas weave, varnish, tempera hatching, watercolour edges and paper grain. Setup: * Model: MOSS-Image SFT (int8), 1024×1536, no upscaling * Same seed for every image * Baseline: 50 steps, CFG 3, skip layer guidance on layer 22, the recommended negative prompt * Turbo LoRAs: strength 1, CFG 1, each at its intended step count (12, 8 or 4) * Each prompt run twice: ER-SDE + simple (top row) and Euler + simple (bottom row) * 5 prompts, one per medium: old-master oil, egg tempera on panel, thin impressionist oil, sepia watercolour over pencil, transparent watercolour Links to each LoRA: 12\_step\_turbo\_moss\_1024\_distill\_v8\_r64\_weightsx2: [https://huggingface.co/anyMODE/moss-image-12step-turbo-lora](https://huggingface.co/anyMODE/moss-image-12step-turbo-lora) HaDeS: [https://huggingface.co/DeVilDoNia/HaDeS-MOSS-Turbo](https://huggingface.co/DeVilDoNia/HaDeS-MOSS-Turbo) Moss Image SFT 8 steps DMD v0.1: [https://civitai.red/models/3006599/moss-image-turbo-loras?modelVersionId=3411385](https://civitai.red/models/3006599/moss-image-turbo-loras?modelVersionId=3411385) Lodestones 8-steps-dmd2-lora: [https://huggingface.co/lodestones/Kromoss](https://huggingface.co/lodestones/Kromoss) Moss Image SFT 4 Steps v2 600 rank 72: [https://civitai.red/models/3006599/moss-image-turbo-loras?modelVersionId=3414934](https://civitai.red/models/3006599/moss-image-turbo-loras?modelVersionId=3414934) Moss SFT 4 Step rank 64: [https://huggingface.co/lvladikov/MOSS-SFT-Distill-4step-LoRA](https://huggingface.co/lvladikov/MOSS-SFT-Distill-4step-LoRA)
document fetched 2026-11-20 00:30 · canonical https://reddit.com/r/StableDiffusion/comments/1y0d74y/9_loras_vs_50step_baseline_comparison_of