I'll be uploading different Moss Image Turbo LoRAs I use here and crediting the original creators. <h3 id="included-loras">Included LoRAs</h3><ul><li>lodestones/MOSS-Image-SFT-8-steps-dmd</li><li>lodestones/MOSS-Image-Base-8-steps-dmd</li><li>lodestones/Kromoss 8-steps-dmd2-lora</li><li>akhaliq/MOSS-Image-SFT-4-steps (v2.0)</li></ul> Intended Use:<ul><li>SFT 8-step: Trained for MOSS-Image-SFT, but can also be used on Base.</li><li>Base 8-step: Trained for MOSS-Image-Base.</li><li>Base 8-step DMD2 (from the Kromoss repo): Distilled from MOSS-Image-Base, then refined with a DMD2 GAN stage against real images. Works on Base, SFT and Kromoss. Gives slightly more contrast and detail than the plain DMD version, and is the creator’s recommended pick.</li><li>SFT 4-step v2: Trained for MOSS-Image-SFT. Euler-native, so it runs on the stock sampler, also can be used on Base.</li></ul><h3 id="recommended-settings">Recommended Settings</h3>Steps:<ul><li>8-step LoRAs (including DMD2): 8 steps. The Base and DMD2 versions also work at 4.</li><li>4-step LoRA: 4 steps up to 1024×1024, 6 steps around 1536 wide, 8 steps at 2048×2048. At 4 steps above 1024², images look hazy.</li></ul>CFG: 1.0 (designed to run without CFG)Sampler: Euler (flow-matching), shift 5LoRA Strength:<ul><li><code>moss_image_sft_8_steps_dmd_v0.1</code>: 1.0</li><li><code>moss_image_base_8_steps_dmd_v0.1</code>: 1.0</li><li><code>moss_image_base_8_steps_dmd2_v0.1</code>: 1.0</li><li><code>moss_image_sft_4step_v2.0</code>: 1.0</li></ul>Use one turbo LoRA at a time.<h3 id="disclaimer-and-attribution">Disclaimer &amp; Attribution</h3>I did not train, distill, or create the original Turbo/DMD LoRA weights. This model page provides modified compatibility versions / repackaged safetensors intended to let compatible LoRA adapters load seamlessly with ComfyUI’s built-in LoRA loader.Original Creators:<ul><li>Full credit goes to <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones">lodestones</a> for the <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd">MOSS-Image-SFT-8-steps-dmd</a> and <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/MOSS-Image-Base-8-steps-dmd">MOSS-Image-Base-8-steps-dmd</a> distillations, and the 8-steps-dmd2 LoRA from <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Kromoss">Kromoss</a>.</li><li>Full credit to <a target="_blank" rel="ugc" href="https://huggingface.co/akhaliq">akhaliq</a> for the <a target="_blank" rel="ugc" href="https://huggingface.co/akhaliq/MOSS-Image-SFT-4-steps">MOSS-Image-SFT-4-steps</a> distillation (warm-started from lodestones’ 8-step work).</li></ul>License: Apache-2.0<h3 id="original-creator-statement-notes">Original Creator Statement / Notes</h3>SFT: “Work-in-progress distillation of OpenMOSS-Team/MOSS-Image-SFT (50 steps, CFG 3) to 8 network evaluations, no CFG. Data-space DMD (arXiv 2610.12102): the generator iterates on its own clean estimate from fixed noise. Trained on compact-JSON scene_graph captions. Checkpoint: update 2600 (EMA), float32 weights.”Base: “Work-in-progress distillation of OpenMOSS-Team/MOSS-Image-Base (50 steps, CFG 3) to 8 network evaluations, no CFG. Data-space DMD (arXiv 2610.12102): the generator iterates on its own clean estimate from fixed noise. Trained on compact-JSON scene_graph captions. Checkpoint: update 12000 (EMA), float32 weights.”Base DMD2 (summary): This LoRA continues the 12,000-update Base data-space DMD run for another 3,000 updates (EMA) using DMD2’s second stage. That stage adds a GAN discriminator trained against real images, at a quarter of the learning rate. It is an fp32 rank-128 LoRA. On the creator’s PartiPrompts CLIP benchmark, it scored highest of the 8-step options on both Base and Kromoss. Keep it unmerged rather than merging it into bf16 weights.SFT 4-step v2.0 (summary): A 4-step, no-CFG distillation of MOSS-Image-SFT. Version 2.0 (released 2026-10-18) is trained directly on the stock flow-matching Euler schedule. It is a single fp32 rank-128 LoRA. The creator reports about 11 s per 1024² image on an A100, accurate English and Chinese text, and no tiling at 2048². For best quality, keep the LoRA unmerged rather than merging it into bf16 weights.

Moss Image Turbo LoRas

moss_image_sft_4_steps_lora_v2_600_avg_rank_72_bf16.safetensors

moss_image_sft_4step_v2.0.safetensors

MOSS_Image_SFT_4_steps_v2

kromoss_8_step_dmd2_v2

MOSS_Base_8_step_dmd_v0.1

Moss_SFT_8_steps_dmd_v0.1

MOSS-Kedama(玉之けだま) Art StyleNo trigger words are required, and the default strength is recommended. It was trained on BASE using SFT and can be used with acceleration LoRAs for image generation.The images include a simple workflow. Just drag and drop them into ComfyUI to load the workflow.The acceleration models are available in the attached files. Download them as needed. Recommended settings: CFG 1, 4–8 steps.

MOSS-Kedama(玉之けだま) Art Style

style-moss-image-base-epoch-07.safetensors

v1.0

Files with SHA256: 154c7f1455c3bf9925945c2f1ea2e4f09d0d0c67256081384a094b2b585f913f