I Built a Brush-Based AI Photo Eraser (People, Objects, Text, Watermarks)

I’m leony, an indie maker. I kept wanting a small, focused flow for cleaning a photo: paint over the distraction, let AI rebuild that patch from the surrounding scene, download, done. Full design suites are overkill for that, and generic “generate a new image” tools don’t preserve the rest of the frame the way a local erase-and-fill does.

So I built RemoveGenie — a brush-based AI photo eraser for people, objects, text, watermarks, and stickers.

How the brush + fill works

The product loop is intentionally short:

1. Upload a JPG, PNG, or WebP (up to 20 MB).
2. Paint a red mask over what you want gone. Undo, Clear, and a brush-size slider are there when the first pass is messy.
3. Sign in with Google and submit when the cost/credits look right.
4. The model reconstructs the masked region from context around it. Covered pixels are generated, not recovered from under the object.
5. Download a PNG. Large inputs are resized so the longest edge is at most 1536 px — check dimensions before anything print-related.

That “generated, not recovered” line matters. If a tourist’s head covered a storefront sign, you get a plausible wall or window, not the real letters that were behind them.

What I use it for

Same editor, six focused tool pages so the intent is obvious:

• People — strangers in a coastal portrait, photobombers on a trip shot (remove people from photo).
• Objects — a cyclist at the edge of a street frame, clutter on a desk (AI object remover).
• Text — burned-in captions, date stamps, a few short words on an image.
• Watermarks — only on images you have the right to edit. Removing a mark does not transfer copyright.
• Stickers / emojis — overlays on screenshots or shared photos.
• Magic eraser — the same brush/fill loop when you just want “erase this blob.”

Include shadows and reflections in the mask when you can; the model does not reliably invent unmarked ones.

Honest limits (so you don’t burn credits)

• Login to submit. You can prepare the mask first, but Google sign-in is required to run a removal.
• 16 credits per removal. New accounts get 16 credits (enough for one try), valid 30 days. Confirmed failed tasks are refunded automatically per the site FAQ.
• PNG out, resized. Expect a PNG download and the 1536 px long-edge cap — not print-archive fidelity.
• Area size. It works best when the erase region is under roughly a third of the frame. Huge masks leave the model with too little context and tend to fail or look muddy.
• Not OCR, not recovery. Text removal does not extract wording for you, and stickers/blurs over faces or copy cannot reveal the real hidden detail — only a plausible fill.
• Checkout. One-time credit packs are listed on the pricing page; as of early October 2026 buy buttons could show checkout unavailable. Don’t plan a paid run until checkout is live for your account.

Try it

Product home: https://removegenie.com/

Owned repo: https://github.com/wayperlee/removegenie

If you clean travel portraits or marketing stills often, I’d love feedback on edge cases — crowded overlaps, tiny wires on busy backgrounds, and dense UI screenshots are the hardest for me right now.