<h1 id="moss-image-sft-deband-(int8)">MOSS-Image SFT DeBand (INT8)</h1>MOSS-Image SFT has a tendency to produce vertical color banding, particularly in skies, gradients, fog, and other areas of relatively uniform color.DeBand addresses this by replacing a small number of transformer layers with their counterparts from MOSS-Image Base. This reduces the banding at the model level rather than trying to correct it through postprocessing.The goal is to preserve the qualities of MOSS-Image SFT while getting rid of an annoying artifact.This release includes three ready-to-use INT8 ConvRot checkpoints for ComfyUI. Each is approximately 5.4 GB, with no additional LoRAs required.<h2 id="versions">Versions</h2><h3 id="deband+">DeBand+</h3>The stronger correction, replacing transformer layers 32–41 with their Base counterparts.This produced the lowest average banding score in our tests and is particularly useful for landscapes, smooth gradients, and wider compositions where the artifact can become more apparent.Recommended settings: 25 steps.<h3 id="deband">DeBand</h3>A more conservative correction, replacing layers 37–41.Retains more of the original SFT weights while still substantially reducing banding. A good starting point if you want results as close as possible to the original model.Recommended settings: 25 steps.<h3 id="deband-fast">DeBand Fast</h3>The DeBand correction combined with lodestones' DMD acceleration, baked directly into the checkpoint.Runs at 10 steps instead of 25, with some trade-off in fine detail.Recommended settings: 10 steps.<h2 id="results">Results</h2>We tested the models across five prompts designed to expose banding in different situations, including skies, fog, and smooth gradients.ModelMean banding score (lower is better)Original MOSS-Image SFT INT82.09DeBand INT81.36DeBand+ INT81.15DeBand Fast INT81.20DeBand+ had the strongest average banding reduction, followed closely by Fast.These are results from a small test set rather than an exhaustive benchmark. The Fast variant also uses a different sampling configuration, so its numbers aren't directly comparable in every respect.Comparison images and additional details are available in the <a target="_blank" rel="ugc" href="https://huggingface.co/MATLOWAI/MOSS-Image-SFT-DeBand-int8?utm_source=chatgpt.com">Hugging Face repository</a>.<h2 id="installation">Installation</h2>Download your preferred checkpoint and place it in:<code>ComfyUI/models/diffusion_models/</code>You'll also need the text encoder and VAE from <a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image">Comfy-Org/MOSS-Image</a>.Text encoder:<code>qwen3.5_9b_int8_convrot.safetensors</code>Place in <code>ComfyUI/models/text_encoders/</code>VAE:<code>flux2-vae.safetensors</code>Place in <code>ComfyUI/models/vae/</code>These are diffusion-model-only checkpoints, so use ComfyUI's Load Diffusion Model node rather than the standard checkpoint loader.<h3 id="comfyui-workflow">ComfyUI Workflow</h3>A ready-to-use workflow is available here:<a target="_blank" rel="ugc" href="https://huggingface.co/MATLOWAI/MOSS-Image-SFT-DeBand-int8/blob/main/workflows/moss_deband_int8.json">Download the MOSS-Image DeBand INT8 workflow</a>Load it into ComfyUI, select your checkpoint, and adjust the sampling settings for the version you're using.<h3 id="sampling-settings">Sampling Settings</h3>SettingDeBand / DeBand+FastSteps2510CFG33SamplerER-SDEER-SDETemporal Score RescalingEnabledEnabledSkip-Layer Guidance22, 2422, 24These are the configurations used for the published comparisons. Other settings may work well too.<h2 id="technical-details">Technical Details</h2>The banding appears to be associated with changes introduced during SFT rather than being inherent to the base model.By comparing the two sets of weights, we found that replacing a relatively small group of later transformer layers could substantially reduce the artifact without reverting the entire model to its Base behavior.DeBand replaces layers 37–41, while DeBand+ extends the replacement to layers 32–41.The INT8 quality variants use the original Comfy-Org quantized weights for the selected layers, avoiding an additional quantization pass.The Fast variant combines the DeBand correction with lodestones' DMD acceleration before quantization.The approach is fairly simple, but the improvement is noticeable, especially in images where broad areas of similar color make the original artifact difficult to ignore.<h2 id="notes">Notes</h2>These models require a recent version of ComfyUI with MOSS-Image and INT8 ConvRot support.The text encoder and VAE are distributed separately. The 5.4 GB checkpoint size doesn't represent total VRAM requirements.DeBand and DeBand+ generally preserve the original SFT's character, though replacing transformer layers can affect generation behavior beyond banding. Fast makes an additional quality/speed trade-off.The correction isn't guaranteed to eliminate every instance of banding, particularly at unusual resolutions or with certain prompts.<h2 id="downloads-and-research">Downloads and Research</h2>INT8 checkpoints, workflows, comparisons, and checksums:<a target="_blank" rel="ugc" href="https://huggingface.co/MATLOWAI/MOSS-Image-SFT-DeBand-int8?utm_source=chatgpt.com">MOSS-Image-SFT-DeBand-int8 on Hugging Face</a>Original research, BF16 variants, and LoRA-based corrections:<a target="_blank" rel="ugc" href="https://huggingface.co/MATLOWAI/MOSS-Image-SFT-DeBand">MOSS-Image-SFT-DeBand on Hugging Face</a><h2 id="credits">Credits</h2><ul><li><a target="_blank" rel="ugc" href="https://huggingface.co/OpenMOSS">OpenMOSS</a> for the original MOSS-Image models.</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Comfy-Org/MOSS-Image">Comfy-Org</a> for the INT8 ConvRot releases and ComfyUI integration.</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/lodestones">lodestones</a> for the DMD acceleration work used in the Fast variant.</li><li><a target="_blank" rel="ugc" href="https://huggingface.co/Kijai">Kijai</a> for the SFT-difference LoRAs and observations that helped motivate the banding investigation.</li></ul>Released under Apache 2.0, subject to the applicable upstream component licenses.

MOSS-Image SFT DeBand — Fix Color Banding | INT8 | Fast & Quality Versions

DeBand INT8 (25 Steps)DeBand is the more conservative banding correction, replacing transformer layers 37–41 with their MOSS-Image Base counterparts.It substantially reduces vertical color banding while staying closer to the original MOSS-Image SFT weights and generation behavior.In our five-prompt evaluation, DeBand achieved a mean banding score of 1.36 compared with 2.09 for the original SFT INT8.This version uses INT8 ConvRot weights, with no additional LoRAs required.Recommended settings: 25 steps, CFG 3, ER-SDE sampler, Temporal Score Rescaling enabled, skip-layer guidance on layers 22 and 24.Requires a compatible MOSS-Image text encoder and VAE. See the main model description for installation instructions and the ComfyUI workflow.

mossImageSFTDebandFixColor_debandINT825Steps.json

mossImageSFTDebandFixColor_debandINT825Steps.safetensors

DeBand INT8 (25 Steps)

DeBand Fast INT8 -10 Step

DeBand+ INT8 (25 Steps)

DeBand+ is the stronger of the two quality-focused banding corrections, replacing transformer layers 32–41 with their MOSS-Image Base counterparts.It achieved the lowest average banding score in our five-prompt evaluation (1.15 versus 2.09 for the original SFT INT8), particularly improving smooth gradients, skies, and backgrounds where vertical stripes are most apparent.This release uses INT8 ConvRot weights, with no additional LoRAs required.Recommended settings: 25 steps, CFG 3, ER-SDE sampler, Temporal Score Rescaling enabled, skip-layer guidance on layers 22 and 24.Requires a compatible MOSS-Image text encoder and VAE. See the main model description for installation instructions and the ComfyUI workflow.

DeBand Fast INT8 (10 Steps)DeBand Fast combines the conservative DeBand correction with reduced lodestones' DMD acceleration leaving less distillation impact, baked directly into the checkpoint.Designed for faster generation at 10 steps instead of 25, with some trade-off in fine detail compared with the quality-focused variants.In our five-prompt evaluation, Fast achieved a mean banding score of 1.20 compared with 2.09 for the original SFT INT8. The accelerated sampling configuration differs from the quality variants, so these scores aren't directly equivalent.This version uses INT8 ConvRot weights, with no additional LoRAs required.Recommended settings: 10 steps, CFG 3, ER-SDE sampler, Temporal Score Rescaling enabled, skip-layer guidance on layers 22 and 24.Requires a compatible MOSS-Image text encoder and VAE. See the main model description for installation instructions and the ComfyUI workflow.

Files with SHA256: 3fb36959a80ce050c3970ea5a0a72e8bd518e1cecd0e492e472cc53cbdfc13dd