<h2 id="original-huggingface-repo-for-model-at-lodestoneskromoss">Original Huggingface repo for model at <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Kromoss">lodestones/Kromoss</a></h2><h3 id="ask-questions-on-the-huggingface-repo-if-anything.no-comments-will-get-responded-to.">Ask questions on the Huggingface repo if anything.<br />No comments will get responded to.</h3><h3></h3><p></p><p>Kromoss is a full fine-tune of <a target="_blank" rel="ugc" href="https://huggingface.co/OpenMOSS-Team/MOSS-Image-Base">MOSS-Image-Base</a> at</p><p>1024 px. It uses the same architecture, inference pipeline, text encoder (Qwen3.5-9B) and VAE (FLUX.2 small</p><p>decoder) as <a target="_blank" rel="ugc" href="https://huggingface.co/OpenMOSS-Team/MOSS-Image-SFT">MOSS-Image-SFT</a>, and this repo has the same</p><p>layout, so it is a drop-in replacement. The fine-tune mainly fixes the Base model's blurriness.</p><p>The repo also ships <a target="_blank" rel="ugc" href="https://huggingface.co/lodestones/Kromoss/blob/main/8-steps-dmd-lora.safetensors">8-steps-dmd-lora.safetensors</a>, an 8-step turbo LoRA (data-space DMD distillation of</p><p>MOSS-Image-Base): <strong>8 steps, CFG 1</strong>, no negative branch.</p>

Files with SHA256: a2e84c9cb86a61b322bb1b75c1ecf02fc5e448176033ef4a82c61eb09f706fbc