File size: 2,547 Bytes
6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 d7f3929 9d2f357 d7f3929 9d2f357 d7f3929 9d2f357 d7f3929 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 6a361e2 9d2f357 9842bd2 9d2f357 9842bd2 9d2f357 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 | ---
language:
- en
- zh
license: apache-2.0
base_model: OpenMOSS-Team/MOSS-Image-SFT
base_model_relation: adapter
pipeline_tag: text-to-image
tags:
- moss-image
- lora
- comfyui
- beta
- cfg1
- 16-steps
---
# 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
- [BF16 LoRA — 595 MB](16step-cfg1/MOSS-Image-16step-cfg1-BF16.safetensors)
- [ComfyUI workflow](16step-cfg1/workflows/MOSS-Image-16step-cfg1-ui.json) · [workflow ZIP](16step-cfg1/MOSS-Image-16step-cfg1-Workflow.zip)
- Base: original [MOSS-Image-SFT](https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT), BF16.
- **16 steps · CFG 1 · Euler / simple · flow shift 5.**
- LoRA strength **1.0**; no trigger; empty negative prompt; no skip-layer guidance.
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](https://github.com/OpenMOSS/MOSS-Image/tree/608158b568fa2e769d918ceccaaf2461f52b3e38/comfyui). 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](16step-cfg1/release.json) and [SHA256SUMS](16step-cfg1/SHA256SUMS).
## Examples
Seven single-pass SFW examples, without upscaling or retouching. [Prompts and seeds](16step-cfg1/examples.json) are included.
Created by Coffmeteor, based on OpenMOSS. Apache 2.0.
|