MOSS Image Turbo Lite — 16step-cfg1 · Beta

Beta acceleration LoRA for MOSS-Image-SFT. 16 steps, CFG 1.

This version can be tried as a fast image-generation option. It provides some improvement in overly oily/glossy skin appearance. Multi-person limb correctness and anatomy have not improved; those problems remain.

Download and settings

This is an experimental beta, not an anatomy correction model. Judge the result for your own prompts and use cases.

ComfyUI setup

Install the official MOSS HF extension. Put the complete original SFT repository in ComfyUI/models/diffusers/MOSS-Image-SFT, and this LoRA in ComfyUI/models/loras/moss-image/. Load the supplied UI workflow. Start ComfyUI from its root or edit the loader's model path.

The workflow uses the HF loader with ModelSamplingSD3 shift 5 and one complete 16-step pass. Keep that loader/sampling-node pairing. The LoRA is stored in BF16, rank 128 / alpha 128; exact file identity is in release.json and SHA256SUMS.

Examples

Seven single-pass SFW examples, without upscaling or retouching. Prompts and seeds are included.

Silver Couture Portrait

Bookshop Portrait

Holiday Studio Portrait

Velvet Lounge Portrait

Jazz Guitarist

Jade Tea Still Life

Moonlit Journeys Poster

Created by Coffmeteor, based on OpenMOSS. Apache 2.0.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Coffmeteor/MOSS-Image-Turbo

Adapter
(43)
this model