MOSS-Image is a 5B-parameter Diffusion Transformer from OpenMOSS, built for strong in-image text rendering and controllable layouts. This is a mirror of Abiray’s INT8 ConvRot quantization of the SFT checkpoint. It is the version that loads cleanly with the comfyui-moss-Image custom node.Base vs SFT<ul><li>Base: Full training signal, recommended starting point for LoRA training.</li><li>SFT (this file): Fine-tuned on curated data. Produces cleaner text grouping and more organised layouts. Recommended for generation.</li></ul>Key strengths<ul><li>Accurate English and Chinese text inside images (up to ~4K token prompts).</li><li>JSON layout control for precise placement of text and objects (abs_centralized + yxyx).</li><li>Works with natural language, tags, or layout JSON.</li><li>Photoreal and stylised output.</li><li>INT8 keeps VRAM usage low while staying close to the BF16 quality.</li></ul>Required companion files You also need these two files:<ul><li>Text encoder: qwen3.5_9b_int8_convrot.safetensors</li><li>VAE: flux2-vae.safetensors</li></ul>Place the diffusion model in models/diffusion_models, the text encoder in models/text_encoders, and the VAE in models/vae.Recommended settings<ul><li>Sampler: Euler</li><li>Scheduler: sgm_uniform</li><li>Steps: 30-50</li><li>CFG: 3.0 (3–6 range)</li><li>Shift: 5.0</li><li>SLG: Uncond skip on layer 22 (24/26 also useful)</li><li>Resolution: multiples of 16, 1–2 MP preferred</li></ul>ComfyUI Works with the custom node by wanaigc (<a target="_blank" rel="ugc" href="https://github.com/wanaigc/comfyui-moss-Image">https://github.com/wanaigc/comfyui-moss-Image</a>) and with native support (ComfyUI ≥ 0.41.3 with the MOSS-Image nodes). Keep SLG on for complex scenes to reduce anatomy and detail collapse.A working sample workflow is attached, and <a rel="ugc" href="/models/3008908/moss-image-sft-int8-multi-lora-daiver-v10-optimized-workflow-for-moss-image-sft-int8-with-multi-lora-seedvr2-upscaling-and-civiscribe">I have published my own refined workflow already</a>.Original models and code: <a target="_blank" rel="ugc" href="https://github.com/OpenMOSS/MOSS-Image">OpenMOSS/MOSS-Image</a>. Quantization by Abiray. Apache 2.0. All credit to the OpenMOSS team.

Files with SHA256: 2191eebd9d7172f51df4427282773915dc6b1cc0193a563c96855f87d9ecaaae