How to Replace an Object in a Photo Without Photoshop

You have a perfect photo. There's one thing wrong with it. A power line through the sky. A coffee cup someone left on the table. A person who walked into the frame at the wrong moment. A logo on the t-shirt you can't legally use.

Old answer: open Photoshop, clone-stamp tool, painstaking patching, 15 minutes minimum.

New answer: brush over the thing, type what you want there instead, wait ten seconds.

This is inpainting — and it's the single most useful AI photo edit feature for everyday use.

The basic flow

1. Open AI Pass Canvas
2. Drop your photo onto the canvas
3. Open the radial menu, tap Inpaint
4. Brush over the area you want changed (the brush is mask-style — paint the region)
5. Type what should be there instead in the prompt box
6. Click Apply
7. Wait ~10 seconds

Only the painted area changes. The rest of the photo is untouched, pixel-perfect.

Examples that work well

• "Remove this thing" — paint over the object, prompt: "empty grass" or "clear sky"
• "Replace this object" — paint over the coffee cup, prompt: "ceramic teapot, same lighting"
• "Fix this person's face" — paint over a closed-eyes face, prompt: "person with eyes open, same expression"
• "Remove a logo" — paint over the t-shirt logo, prompt: "plain navy shirt fabric"
• "Change the sky" — paint the sky region, prompt: "dramatic sunset clouds, golden hour"
• "Take down a power line" — paint over the line, prompt: "open sky, no obstructions"

What model is doing the work

Inpaint in Canvas uses Nano Banana 2 by default (Gemini 3.1 Flash Image, $0.067/edit) — the model is mask-aware, meaning it knows to only modify what you brushed and to blend the new content seamlessly with the surrounding photo.

You can also pick GPT Image 1 ($0.04/edit) from the model dropdown — it's slightly cheaper but the prompt-following on inpaint specifically is a step behind Nano Banana 2.

Tips for getting good results first try

Be specific in the prompt. "Remove the cup" works less reliably than "empty wooden table surface, same lighting." The model needs to know what to put there, not just what to remove.

Mention lighting and context. "...same lighting," "...same color palette," "...consistent with rest of image" — these prompts produce more seamless blends.

Brush slightly outside the object. Including a few pixels of surrounding background gives the model more context and tends to produce cleaner edges than tight masks.

For object removal, prompt the background. Don't say "no person." Say what should be visible if the person weren't there: "wooden bench, autumn leaves, same daylight."

Where it still struggles

• Removing things from busy patterned backgrounds (bookshelves, brick walls with specific texture). The model fakes it but a careful eye can spot the seam.
• Multiple complex changes in one mask. Better to do them as separate inpaints.
• Faces of recognizable specific people. Models are conservative on identity — "make this person smile" works less well than you'd hope on close-ups.

For those cases, Photoshop content-aware fill or manual cloning still beat AI inpainting. For everything else — and that's most of everything — inpaint is the answer.

What it costs

Roughly 7 cents per inpaint with Nano Banana 2. Your $1 signup credit covers about 14 of them. If you only inpaint occasionally — which most of us do — that's a year of edits before you'd top up.

Try it

AI Pass Canvas is free to open. Paste a photo, tap Inpaint in the radial menu, brush over the bit you want changed, write a prompt. The first 14 attempts are on the house.