PixelPanda: Remove Text/Watermark

PixelPanda: Remove Text/Watermark is the most "just make it go away" node in the ComfyUI-PixelPanda pack. You feed it an image that has a caption, a timestamp, a logo or a watermark sitting on top of it, it calls the PixelPanda API's /remove-text endpoint, and the offending text comes back gone - 1 credit per image, zero parameters, no prompt, no mask, no telling it which text you meant. It finds the text and inpaints over it automatically.

How it works

Same architecture as the rest of the pack: the image goes out as base64 PNG to https://pixelpanda.ai/api/v2, the hosted service detects the text regions and regenerates what's underneath them, and the result comes back as an IMAGE in your graph. Text removal is a classic inpainting job, and one the modern instruction-editing wave handles well - the KB's inpainting doc notes that "remove this object" is exactly the work that moved from masked inpaints to full-frame edit models. This node collapses all of that - detection plus regeneration - into one opaque cloud call. What you give up for that convenience is control: no way to draw a box around one label and spare another, no strength dial, no re-run of just the failed region. If the service decides your poster text is worth removing and your graphic design text isn't, that's the call it makes.

The inputs

• image - the offending image.
• api_key - pk_live_… key, or leave blank to fall back on PIXELPANDA_API_KEY.

One IMAGE out. That's genuinely all there is to it - this node has no settings to get wrong, which is either its best feature or its most frustrating one depending on whether the auto-detection agrees with you.

Install and key

cd ComfyUI/custom_nodes
git clone https://github.com/RyanKramer/ComfyUI-PixelPanda

Restart ComfyUI and it lives under image/PixelPanda; or install via ComfyUI Manager by searching "PixelPanda". No dependencies to add - the pack's requirements are empty and it leans on the torch/numpy/Pillow already in ComfyUI. Get a key from https://pixelpanda.ai/profile, drop it in the node or export it as PIXELPANDA_API_KEY before launch. The $1.99 starter pack needs no subscription and this node burns 1 credit per image.

Where this fits - and the uncomfortable part

The clean use case is cleaning up your own renders: killing stray watermark text baked into a generation, removing a timestamp you didn't want, scrubbing a caption off a mood board you made. If you generate for print-on-demand, this is the pre-upload cleanup pass that used to cost you a manual inpaint. The uncomfortable part is the same one that applies to any automatic watermark remover: stripping someone else's watermark off their image to pass it off as yours is not a technical problem, it's a copyright one, and the KB's background-removal and inpainting docs carry the same warning about shipping other people's work through your store. The tool doesn't judge; you should.

Beyond that, set expectations: without a mask or prompt, "remove this text but keep that one" is not something this node understands, and text sitting over busy texture is where inpainting-based removers leave artifacts that are obvious at a glance. When the auto-detection nails it, it's a magic one-node fix. When it half-misses, you'll wish it had a mask input - at which point a local inpaint of the specific region is the honest fallback.

Gotchas

Only the first frame of any batch gets processed, so no batching video frames. The call needs the network and can take up to 300 seconds; the console line [PixelPanda] /remove-text ok tells you it finished. Bad key reads 401, no credits reads 402, both with clear messages. And the standing caveat for this whole pack applies: PixelPanda is a small new service with a thin community footprint, so if the API goes away the node goes quiet with it. Fine for occasional cleanup, not something to bet a pipeline on.