Bria Eraser
Remove anything under a mask and let the cloud fill the hole — Bria Eraser
- image
- mask
- IMAGE
The photobomber problem, solved by one mask and an API call. Bria Eraser takes an image, takes a mask covering whatever you want gone, and returns the same image with that thing removed - the hole filled in with plausible content that matches the surroundings. No prompt needed, no local model, no inpainting pipeline to babysit.
It sits in the BRIA AI API nodes pack, and it's the most literal of the mask-based editing nodes: image + mask in, image out, everything handled on BRIA's servers at engine.prod.bria-api.com. You need a BRIA API token from platform.bria.ai, and you don't need a single byte of VRAM. BRIA's models train on 100% licensed data, which is the reason this exact node keeps showing up in commercial retouching workflows where nobody wants to explain a scraped-dataset provenance question to legal.
How it works
The image and mask get base64-encoded and POSTed to BRIA's /v2/image/edit/erase endpoint. The API returns a status_url, the node polls it until the job says COMPLETED, then downloads the result and returns it as an IMAGE tensor - ready for a Save Image node or anything else downstream. The whole pack shares this poll-until-done machinery; Eraser is just the simplest instance of it, with the fewest knobs.
The inputs that matter
image,mask, andapi_key- that's the whole required set.- The
maskis a standard ComfyUI MASK. Paint it by hand, or generate it - segment a person and feed the mask in. It gets downscaled/encoded on your side, so paint generously but tightly: the fill quality tracks your mask edge quality, exactly like local inpainting does. - Two optional content-moderation toggles (
visual_input_...andvisual_output_...), both off by default.
That's genuinely all there is. Eraser is the "no decisions" node - which is both its strength and its limit. There's no prompt to tune a fill's content; the API just makes the hole disappear as naturally as it can.
Install
Standard pack install - nothing to compile, no requirements file, no models to download:
cd path_to_comfyui/custom_nodes
git clone https://github.com/Bria-AI/ComfyUI-BRIA-API.git
or search "BRIA API" in ComfyUI Manager and install, then restart ComfyUI and paste your token into api_key.
Eraser vs. GenFill - get this right
They take identical inputs and live side by side in the same pack, which guarantees confusion. Eraser removes what's under the mask and fills the gap with background (no prompt). Bria GenFill also takes image+mask, but requires a prompt and generates a new object inside the mask. Photobomber → Eraser. Photobomber replaced by a plant → GenFill. If a node demands a prompt field, you're on the fill side; if it doesn't, you're erasing.
One thing to budget for: every execution is a paid API call, and erasing a large region can produce a fill you want to retry. Set a seed? You can't here - Eraser has no seed input, so retry by rerunning or by nudging the mask. That's the one real frustration with this otherwise dead-simple node.
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| image | IMAGE | — | |
| mask | MASK | White areas are erased, black areas are preserved. The mask is binarized before sending, so partially painted areas count as white. Must have the same aspect ratio as the image. | |
| mask_type | COMBO | manual for hand-drawn or brush masks, automatic for masks produced by segmentation models such as SAM. | |
| moderation | COMBO | Moderation settings |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| IMAGE | IMAGE | — |