To remove an object from a photo with AI4Chat, upload the picture to the photo editor, type something like “Remove the man in the red jacket on the left. Fill the space with the same brick wall,” choose Nano Banana and generate. You do not paint over anything. Google lists “remove an entire person from a photo” and “remove a stain in a t-shirt” among the edits it built Nano Banana for, and for the objects that are hard to put into words there is a second route: draw a bright box around the thing on a copy of the photo and ask the model to remove what is inside it.
How to remove an object from a photo with a prompt
Erase an object with the AI photo editor
- 1
Upload the photo
Open the photo editor and add one picture. Keep the original somewhere safe; you will want it for comparison.
- 2
Write a three-part prompt
What to remove, where it is, and what should fill the gap. “Remove the orange traffic cone in the bottom right. Fill it with the same gray pavement.”
- 3
Choose Nano Banana and generate
If the object survives or something else changes, rerun the same prompt on Flux Kontext before you rewrite it. The editor can pass a very short request to a lighter model, one more reason to write all three parts.
- 4
Check the edges, then download
Zoom in where the object was. Look for leftover shadows, a doubled edge or a pattern that does not line up.
Name it, place it, say what goes behind it
Removal prompts fail in two predictable ways. Either the model cannot tell which of three people you mean, or it removes the right one and fills the hole with something odd. Naming a color or a position fixes the first. Saying what belongs there fixes the second. Ask for shadows and reflections by name, because the model treats them as separate things and often leaves them behind.
Remove the person in the blue shirt walking behind the couple. Fill the space with the same beach and sea. Keep the couple exactly as they are. Remove the power lines crossing the sky. Keep the trees and rooftops unchanged. Remove the trash bin by the front door and its shadow. Fill the area with the same stone steps. Remove the coffee stain on the white tablecloth. Keep the folds and the plates where they are. Remove the scaffolding from the front of the building. Show the building's brick facade and windows behind it. Remove the tourist on the left and his reflection in the water.
Swap in your own object, color and location. Keep the last sentence that says what stays.
When words are not enough, draw a box
Some things are hard to describe: one of twenty identical chairs, a blurry shape at the edge, a sticker on a laptop. For those, mark the photo instead. Open a copy in your phone's markup tool, draw a bright red rectangle around the object, save it and upload that copy.
Two model makers document this. Black Forest Labs' Kontext editing docs describe using bright colored boxes as annotations for targeted local edits and note that the boxes are removed in the output. ByteDance shows the same idea on its Seedream 4.5 page with the prompt “Only keep the character in the green outline, and delete the other characters.” So start with Flux Kontext for boxed edits and try Seedream 4.5 second.
Remove the object inside the red box, and remove the red box too. Fill the area with the surrounding [grass/floor/wall] so it matches the rest of the photo. Change nothing outside the box.
What is easy to remove and what is not
A rule of thumb from how these models work rather than a measured success rate: the less the model has to invent, the better the result. Removing a small object from a plain wall asks it to continue a pattern it can see. Removing a crowd from a famous square asks it to make up half a square.
| What you are removing | What to expect, and what to add |
|---|---|
| A small object on a plain surface | The easiest case: the model continues a surface it can see. Say what the surface is. |
| One person in a busy background | The model has to invent what was behind them. Say what belongs there: “the same crowd”, “empty pavement”. |
| Wires, poles, a stray branch | Thin lines can leave faint traces. Add “and any trace of them”. |
| Something touching your subject | The subject can change too. Add “keep her hand and sleeve exactly as they are”. |
| Half the frame, like a parked bus | The fill is a guess. Describe the scene you expect, then check it. |
| Shadows and reflections | Often left behind. Name them: “and its shadow”, “and his reflection”. |
Prompt removal or a brush eraser?
Brush tools such as Cleanup.pictures work the other way round: you paint over the object and the AI fills the painted area. That is faster and more exact for tiny spots, like dust on a product shot or a blemish. Describing the object wins when the removal is large, when you also want to relight or restyle the photo in the same pass, or when you would rather not trace a fiddly outline with a finger on a phone screen.
| AI4Chat photo editor | Brush eraser (Cleanup.pictures) | |
|---|---|---|
| How you mark the object | Describe it, or draw a box on a copy | Paint over it |
| Best at | Larger removals and scene rebuilds | Small, precise spots |
| Free use | Account needed; edits run on the Unlimited plan, after a 7-day free trial | Free, export limited to 720px |
| Paid price | Unlimited plan, price set by country | $5 a month or $36 a year |
Source: Cleanup.pictures FAQ, checked November 7, 2026; AI4Chat pricing
If what you actually want gone is the whole background, use the background remover instead; it cuts the subject out cleanly rather than repainting. For more removal and cleanup wording, see the photo prompt library, and for model choice in general, the AI photo editor guide.
Questions people ask
Can I remove a person from a photo with AI?
Yes. Identify them by clothing and position (“the man in the gray hoodie on the far right”), say what should be behind them, and tell the model to keep everyone else unchanged. If two people overlap, removing the one in front is harder because the model has to rebuild the one behind.
Why did nothing change after I asked it to remove something?
The model probably could not tell which object you meant. Add a color and a position, or draw a box around it on a copy and use the box prompt with Flux Kontext.
Can it remove text from a photo?
Yes, from your own photos: a street sign you do not want, a name on a mug, a date stamp. Ask it to remove the text and keep the surface underneath. Do not use it to strip watermarks or credits from images that belong to someone else.
Does it work on a phone?
Yes, in your phone's browser. Upload from your photo library, and if you need the box trick, draw the box with the built-in markup tool first and save it as a copy.
Is the object remover free?
You need a free AI4Chat account to open the editor, but an edit costs more than its starting balance, so removals run on AI4Chat Unlimited, which starts with a 7-day free trial; the price for your country is shown before you pay.