LayerMask: Segformer Fashion Setting

This one confuses people at first because it doesn't output a mask - it outputs a setting. It's the checklist you wire into LayerStyle's Segformer clothes-parsing node to tell it exactly which garments and accessories to include in the mask. Think of it as the control panel: 45 toggles, one per clothing category (shirt, jacket, pants, dress, shoe, hat, glasses, tie, all the way down to buttons, bows, and sequins), and it bundles your picks into a single config the segmenter consumes.

Why split it out into its own node? Because fashion segmentation covers a lot of classes, and cramming 45 booleans onto the actual parsing node would be a monster. Separating the selection lets you reuse the same setting across graphs and keeps the parser node clean. This is the standard "auto-mask the clothes" building block for virtual try-on, garment recoloring, and outfit inpainting.

How it works

The Segformer clothes model classifies every pixel of a person into fashion categories. This node doesn't run the model - it builds the filter that says "keep these categories, ignore the rest." You tick the parts you care about, and the downstream Segformer node uses that to produce a mask covering only those garments.

The inputs and outputs that matter

There are no image or mask inputs - it's all BOOLEAN toggles, 45 of them, default false. You turn on the ones you want masked. A few you'll actually reach for:

• shirt, top, jacket, coat, dress, pants, skirt - the main garments.
• shoe, hat, glasses, bagwallet - common accessories.
• The long tail (collar, sleeve, pocket, zipper, bow, sequin, etc.) is for fine-grained selection when you need just a detail.

Leave everything you don't want at false. Output is a single segformer_fashion_setting of type LS_SEGFORMER_SETTING - a pack-internal config type - which plugs into the matching Segformer fashion parsing node. It only connects there.

How to install it

ComfyUI Manager: search ComfyUI Layer Style, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/chflame163/ComfyUI_LayerStyle

Restart afterward. Important: the Segformer parsing node that consumes this setting needs a model - the segformer clothes/fashion weights, which download automatically to the pack's models folder on first run (so the first execution stalls while it fetches). The model-dependent Segformer nodes live in the heavier half of this project; if Manager offers ComfyUI_LayerStyle_Advanced, that's where the parsing side ships, since the pack was split to keep model-dependent nodes separate from the lightweight ones.

Common issues & troubleshooting

The big pack-wide failure is IMPORT FAILED at startup, and the Segformer side is especially prone to it because it pulls in transformers and friends - a broken transformers/tensorflow combo (a very common traceback for this exact pack) will take it down. Fixes: reinstall the pack clean, and make sure you're not running a stale combined install from before the ComfyUI_LayerStyle / ComfyUI_LayerStyle_Advanced split. If only the Segformer nodes fail while the rest load, it's almost always the dependency/model side, not the base pack.

Node-specific: this setting connects only to the Segformer fashion parsing node - nothing else accepts LS_SEGFORMER_SETTING, so if it won't connect, you're aiming at the wrong node. If your mask comes back empty, you probably left every toggle at false; turn on at least the garment you want. And accuracy is the model's, not yours - Segformer clothes parsing is solid on clear full-body shots and gets shaky on occlusion, unusual poses, or busy patterns.