☁️BizyAir HED Soft-Edge Lines
The forgiving ControlNet edge detector
- image
- IMAGE
HED - Holistically-Nested Edge Detection - produces soft, gradient edges with varying thickness rather than Canny's hard binary lines. It's the forgiving option in the edge-preprocessor family: good on organic subjects like skin, fur, foliage, and fabric, where a perfectly crisp edge map doesn't really exist and Canny would just pick up noise instead of structure.
How it works
The wider ecosystem still describes HED (now often labeled "soft edge" on modern union ControlNets) exactly this way: "produces gradient edges with varying thickness. More forgiving than Canny - doesn't require perfect edge detection. Good for organic subjects." That forgiveness is the whole reason to reach for it over Canny on portraits, animals, or anything with soft, irregular boundaries. It's also the raw signal the pack's sibling Fake Scribble node repurposes into a looser, scribble-style condition - same detector, different downstream processing.
The inputs and outputs that matter
image- required, the source photo.safe(enable/disable, default enable) - clips extreme edge-strength values so JPEG-compression artifacts or sensor noise in the source photo don't get read as real structure. Leave it on unless you specifically want the rawest possible signal and are working from a clean, high-quality source.resolution(default 512, 64-1024) - internal processing resolution before the edge map is produced.
Output: a single IMAGE - the soft-edge map, ready for a ControlNet that supports soft-edge/HED conditioning.
How to install it
ComfyUI Manager: search BizyAir, install, restart. Or:
cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAir
then restart. BizyAir wants an account and API key set up on first use per the README.
Common issues & troubleshooting
ControlNet barely reacts to the edge map. Confirm the ControlNet checkpoint you loaded downstream actually lists soft-edge/HED among its supported conditions - on current union models this condition is usually one selectable mode rather than a separate file, so double-check the model card before assuming the preprocessor output is at fault.
Edges look noisy or exaggerated on a low-quality source photo. Keep safe enabled - this is precisely the artifact it's meant to suppress.
Wanting tighter, more literal structure-following. HED's whole design intent is looser than Canny - if you actually want hard-edge fidelity, use a Canny preprocessor instead rather than fighting HED's inherent softness with settings.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| safeopt | COMBO | enable | 2 options: enable, disable |
| resolutionopt | INT | 51264–1024 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |