ComfyUI FASHN Human Parser
ComfyUI custom nodes for FASHN Human Parser — a SegFormer-B4 model fine-tuned for human parsing with 18 semantic classes, optimised for fashion and virtual try-on workflows.
Installation
Option A – ComfyUI Manager
Search for FASHN Human Parser and install from the Manager.
Option B – Manual
cd ComfyUI/custom_nodes
git clone https://github.com/<your-fork>/comfyui-fashn-human-parser
cd comfyui-fashn-human-parser
pip install -r requirements.txt
Model weights (~244 MB) are downloaded automatically from HuggingFace on first use.
Nodes
🟦 FASHN Human Parser
Full segmentation node.
| Input | Type | Description |
|---|---|---|
image |
IMAGE | Input image |
device |
combo | auto / cuda / cpu / mps |
dtype |
combo | float32 / float16 / bfloat16 |
model_id |
STRING | HuggingFace model ID (default: fashn-ai/fashn-human-parser) |
| Output | Type | Description |
|---|---|---|
segmentation_image |
IMAGE | Colour-coded visualisation of all 18 classes |
label_id_mask |
MASK | Greyscale mask where pixel value encodes class ID (normalised 0-1) |
🟩 FASHN Human Parser (Mask)
Extracts a binary mask for one or more body-part labels.
| Input | Type | Description |
|---|---|---|
image |
IMAGE | Input image |
label |
combo | Primary label to extract (see table below) |
device / dtype / model_id |
— | Same as above |
extra_labels (optional) |
STRING | Comma-separated additional labels to OR-merge, e.g. dress,skirt |
blur_radius (optional) |
INT | Gaussian blur radius for soft edges (0 = off) |
invert (optional) |
BOOLEAN | Invert the output mask |
Output: mask (MASK)
🟨 FASHN Human Parser (All Masks)
Returns all 18 class masks stacked as a single (18, H, W) MASK batch.
Useful for downstream per-class compositing nodes.
| Output | Type | Description |
|---|---|---|
all_masks |
MASK | 18-frame mask batch (one frame per class, in ID order) |
Label Classes
| ID | Label | ID | Label |
|---|---|---|---|
| 0 | background | 9 | hat |
| 1 | face | 10 | scarf |
| 2 | hair | 11 | glasses |
| 3 | top | 12 | arms |
| 4 | dress | 13 | hands |
| 5 | skirt | 14 | legs |
| 6 | pants | 15 | feet |
| 7 | belt | 16 | torso |
| 8 | bag | 17 | jewelry |
Model Details
- Architecture: SegFormer-B4 (MIT-B4 encoder + MLP decoder)
- Input size: 384 × 576 (W × H)
- Classes: 18 semantic human-body parts
- Base model: nvidia/mit-b4
- License: NVIDIA SegFormer Source Code License
Typical Workflow
LoadImage → FASHN Human Parser (Mask) [label=top, blur_radius=4]
│
▼ mask
InpaintModelConditioning / IPAdapter / ...