Image Watermark Remover

Watermark removal is one of those jobs that used to mean a fiddly local stack: load Flux Kontext dev, wire an inpainting mask, hope the license gods don't frown. BFL's newer Klein line folded editing into the modern Flux generation stack, and this node is the hosted result - a dedicated flux-klein/watermarker-remover/image-to-image channel where you feed one image and get back the same image with the watermark scrubbed. No model files, no VRAM, no masking.

Let's be clear about what "watermark remover" means here and how the community actually uses it. Kontext earned its reputation as "shockingly effective" at removing watermarks - the kind of tool you run on images you have the rights to clean up, or on your own renders that got a platform stamp slapped on them. Same applies to this channel. It's an image-to-image edit: it sees the whole picture and regenerates the affected region to blend, which is a fundamentally different approach from crop-and-heal.

Inputs and outputs

• image - required, up to 10 MB. That cap matters: the model wants to see the full frame to reconstruct texture, so don't crop the watermark out yourself first and expect magic - feed it the whole image.
• No prompt, no strength slider, no resolution knob. This is a single-purpose endpoint, and that's the appeal.

Outputs are the standard triple: image (the cleaned result), request_id, and urls (JSON string of output file URLs). Wire the image output into a Save Image or a preview and you're done.

How it works

Like every BizyTRD_*_API node, this one is generated from BizyAir's model registry. The pack serializes your image, submits it to the hosted endpoint, polls until the edit finishes, and downloads the result back into your graph. The heavy lifting happens on SiliconFlow's infrastructure - you're paying cloud credits for the privilege of not owning a GPU big enough to run Klein locally. The API key comes from BizyAirPlus settings or BIZYAIR_API_KEY unless you override it with a settings node.

Where people get burned

• Overlapping/busy watermarks: a small logo in a corner vanishes cleanly. A large semi-transparent brand mark over textured fabric or foliage can leave a ghost. If the result has a smudge, run it through once more - I2I edits on the same spot usually tighten up.
• The 10 MB cap: compress before you send. JPEG for photos, careful with giant PNGs.
• Failed runs return a red placeholder with error text if you have skip_error enabled - check the urls output for any original URLs before assuming total failure.

Install

It ships with the whole pack:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI, turn BizyAirPlus ON, and confirm your key. Then weigh it against the free option: if you already have a local editing model, this cloud channel is a convenience, not a capability upgrade. Where it genuinely shines is the zero-setup path - no 20 GB download, no licensing fine print to read, just a node that does the one thing.