Fashion Segment (RMBG)

Fashion Segment is Clothes Segment with a much finer wardrobe. Where the basic clothes node knows "Upper-clothes" and "Pants," this one distinguishes a coat from a jacket from a cardigan from a vest, a jumpsuit from a dress, shorts from a skirt. If you're doing serious garment editing - fashion mockups, try-on style workflows, recoloring one specific layer of an outfit - this is the node with the vocabulary for it.

It runs on a SegFormer fashion-parsing model, trained on a detailed fashion dataset with far more clothing classes than a general human parser. So, like the rest of this family, there's no prompt: you tick the garment types you want and it returns a merged mask of them, because the model was trained to recognize each category by name. The tradeoff for that precision is that the model is a bit heavier and the category list is long.

The inputs are the garment toggles: coat, jacket, cardigan, vest, sweater, hood, "shirt, blouse", "top, t-shirt, sweatshirt", sleeve, dress, jumpsuit, cape, pants, shorts, skirt, "tights, stockings", sock, and shoe. Tick any combination and they merge. Some entries are grouped on purpose - "shirt, blouse" is one toggle covering both, because the underlying model treats them as one class. The real settings are the same as the other parsers: process_res (128–2048, default 512) for detail vs. memory, mask_blur and mask_offset for edge cleanup, invert_output to flip, and background / background_color for the output style. Outputs are the standard IMAGE, MASK, and MASK_IMAGE; MASK is what you feed to an inpaint or recolor.

There's one input that surprises people: accessories_options. This is an input port, not a checkbox - it accepts the output of the separate Accessories Segment node. The fashion parser also knows accessories (belts, bags, collars, zippers, and a long tail of trim), but rather than cram fifty toggles into one node, the pack splits them: you configure which accessories you want on the Accessories Segment node, and pipe its accessories_options output into this input to fold them into the same mask. If you only care about garments, leave that port unconnected and ignore it.

Install is the pack routine: "Comfyui-RMBG" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/1038lab/ComfyUI-RMBG, then pip install -r requirements.txt and restart. The fashion parser auto-downloads on first use into ComfyUI/models/RMBG/segformer_fashion/ and caches after.

Practical notes. Because it splits garments so finely, it can also miss finely - a hoodie might land under "top, t-shirt, sweatshirt" plus "hood" rather than one clean box, so if a mask comes back partial, tick the neighboring categories too and let them merge. The grouped labels ("shirt, blouse") are a hint about how the model thinks: it lumps visually similar garments, so don't expect it to tell a blouse from a shirt. Bump process_res for detailed fabrics and add a little mask_blur before inpainting so the recolored garment doesn't show a hard seam.

When to use which: for quick, coarse outfit work - "mask the top," "mask the bottoms" - Clothes Segment is faster and has fewer boxes to think about. Reach for Fashion Segment when the distinction between garment types actually matters to your edit, or when you need to fold in accessories. For a whole-person cutout, neither - that's an RMBG or BiRefNet job.