comfyui-pixelcut

ComfyUI-Pixelcut

ComfyUI custom nodes for the Pixelcut image API.

Background removal — Pixelcut's flagship cutout model, exposed as a native ComfyUI node that takes an IMAGE and returns the cutout plus its alpha MASK.

Install from the Comfy Registry or via ComfyUI Manager.

Nodes

| Node | Inputs | Outputs | | --- | --- | --- | | Pixelcut Remove Background | image, and optional api_key, crop, shadow, shadow_opacity, invert_mask | image (RGB cutout), mask (subject alpha) | | Pixelcut Credits | optional api_key | credits_remaining (INT) |

The mask marks the subject by default (1 = subject). Flip invert_mask to mark the background instead, matching ComfyUI's LoadImage alpha convention.

Installation

Via ComfyUI Manager: search for "Pixelcut" and install.

Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/pixelcutai/comfyui-pixelcut
cd comfyui-pixelcut
pip install -r requirements.txt

Restart ComfyUI. The nodes appear under the Pixelcut category.

API key

Get a key from your Pixelcut API dashboard.

Provide it either way:

• Environment variable (recommended): set PIXELCUT_API_KEY before launching ComfyUI. It never touches your saved workflows.

  export PIXELCUT_API_KEY="sk-..."
  

• api_key widget: type it into the node. Convenient, but note the key is saved into the workflow JSON and embedded in saved-image metadata — don't share those files.

The widget value takes precedence over the environment variable when set.

Usage

Load Image → Pixelcut Remove Background → Preview Image. The mask output can drive compositing, Set Latent Noise Mask, or any mask-consuming node.

Batches are supported (one API call per frame). When crop is enabled, frames may come back at different sizes and can't be stacked into one batch — process such images individually.

See example_workflows/ for a ready-to-load graph.

Credits & options

• Background removal costs credits per call; shadow adds +3 credits per image. Check your balance with the Pixelcut Credits node.
• Image limits: 25 MB, 64×64 min, 6000×6000 max.

How it works

ComfyUI images are local tensors, so the node uploads image bytes directly via multipart/form-data and requests the raw binary result (Accept: image/*) — no intermediate URL hosting or second download. The returned PNG's alpha channel becomes the ComfyUI MASK.

License

MIT — see LICENSE.