# AI photo editor: 8 edits with one prompt each — tested on a single photo

> An AI photo editor changes a photo from a plain-text instruction: it swaps the background, removes a passer-by, changes an outfit, turns summer into winter or rewrites a sign. We made 8 edits to one photo in Nano Banana Pro — the face and pose never changed — and compared four models on the hardest edit, replacing text on a sign. Below are the results and the prompt for every edit.

Published 2026-10-12 · Updated 2026-10-12
URL: https://ai2face.com/blog/izmenit-foto-nejroset

Today you can edit a photo with AI in a single sentence: a generative model reads the image, understands where the person, background, light and text are, and redraws only what you asked for. We put this to the test on one street photo — eight different edits in a row — and separately compared four models on replacing a sign. Every image below was made in the AI2Face generator.

## What an AI photo editor can do

Today's editing models — [Nano Banana Pro](/models/nano-banana-pro), [Nano Banana 2](/models/nano-banana-2), [GPT Image 2.5](/models/gpt-image), [Seedream 5.0](/models/seedream) — work like a photo editor you talk to. You don't select areas or pick brushes: you describe the result in words and the model finds the right part of the image itself.

| Edit | Example instruction | How hard for the model |
|---|---|---|
| Background | “replace the street with a seaside promenade” | low |
| Removing objects | “remove the passer-by and the trash bin” | low |
| Clothing | “dress him in a knitted sweater” | medium |
| Light and time of day | “make it a warm evening, turn the street lamps on” | medium |
| Season and weather | “make it winter with snow” | medium |
| Adding objects | “put a dog next to him” | medium |
| Style | “redraw it as an anime watercolor” | low |
| Text in the photo | “replace the text on the sign” | high |

## The source photo

We generated a typical phone snapshot: a man at a table outside a street café, a green awning above him with the sign «COFFEE SHOP», a passer-by and a grey trash bin behind him. The shot has everything people usually want to change: a person, clothing, background, unwanted objects, light and text.

![A bearded man in a navy blazer sits at a metal table outside a café with a green awning and the sign «COFFEE SHOP», holding a white cup](https://ai2face.com/_astro/source-en.BsJbVb-v_Z2ikIAH.jpg "The source for every edit: an overcast day, a passer-by on the left, a trash bin by the door. Made in Nano Banana Pro; the person is generated for the test")

We made each edit separately, from this same source, in Nano Banana Pro at 2K. Every prompt with a person in the shot got the same safety line: “Keep the man's face, expression, pose, hands, the cup and the composition exactly the same.”

## 8 edits to one photo: results and prompts

### 1. Change the photo background with AI

```
Replace the background behind the man: instead of the city street and café, put
him at the same small table on a sunny seaside promenade with the sea, a white
railing and a clear blue sky behind him. Remove the awning and the sign. Match
the lighting on him to bright sunny daylight.
```

![The same man at the same table, now on a sunny seaside promenade with white railings](https://ai2face.com/_astro/bg-en.COO0YGqC_1Uh03r.jpg "The whole background replaced, the light on his face is now sunny, the pose and cup are unchanged")

The model recalculated the light instead of just cutting the man out: sunny highlights appeared on his face and the shadows got harder. If you need a background for a product card or an ID photo, ask for a plain one: “replace the background with an even light grey.”

### 2. Remove unwanted objects

```
Remove the walking passer-by on the left and the grey metal trash bin by the
café door. Fill those areas naturally with the sidewalk, street and café wall.
Change nothing else in the photo.
```

![The café shot with no passer-by on the sidewalk and no trash bin by the door](https://ai2face.com/_astro/remove-en.C5NEUIm5_Z7xVbf.jpg "The passer-by and the bin are gone; the sidewalk and wall are filled in without a trace")

Removal is the most reliable edit: there's little room for the model to go wrong. Name the object precisely and by its position — “the passer-by on the left,” “the bin by the door” — or the model may remove the wrong thing.

### 3. Change the outfit

```
Change only his clothes: replace the navy blazer and white t-shirt with a chunky
cream knitted wool sweater with a crew neck.
```

![The same man outside the café in a cream knitted sweater instead of a blazer](https://ai2face.com/_astro/clothes-en.CgaZTd6v_Z15I67C.jpg "The blazer became a sweater; the hands and cup stayed exactly where they were")

The fabric follows the pose: the sweater bunches at the elbows and the sleeves cover the wrists. For more on virtual try-ons, see our guide to [AI clothing on a model](/blog/odezhda-na-modeli-nejroset).

### 4. Change the time of day

```
Turn the scene into a warm evening at dusk: deep blue sky, street lamps switched
on, the café windows glowing warm yellow, soft warm light on his face from the café.
```

![An evening version of the shot: a blue sky, glowing street lamps and warm yellow café windows](https://ai2face.com/_astro/evening-en.BkWiBFhL_Z2f9iEb.jpg "Evening: the street lamps and windows lit up, and warm light from the café falls on his face")

Light is what separates an AI edit from a sticker slapped on in an editor. The model changed not only the sky but also the skin tone, the reflections on the table and the shadows.

### 5. Turn summer into winter

```
Turn the scene into winter: fresh snow on the sidewalk, on the awning and on the
table, light snowfall in the air, cold overcast light. Dress him in a dark charcoal
wool coat with a grey knitted scarf instead of the blazer.
```

![A winter version: snow on the sidewalk and table, the man in a grey coat and scarf](https://ai2face.com/_astro/winter-en.CCJ2Czp__iHrTk.jpg "Winter in one instruction: snow, a coat and a scarf. Some snow even settled on the sign")

One instruction changed the season, the weather and the outfit at once. Edits like this are handy for turning a single shoot into content for the whole year.

### 6. Add an object to the shot

```
Add a friendly golden retriever sitting on the sidewalk next to his chair, looking
up at him, with a natural shadow on the pavement that matches the overcast light.
```

![A golden retriever sits by the table next to the man and looks up](https://ai2face.com/_astro/add-en.CJ1JHwTf_1brSp4.jpg "The dog matches the scale and light of the scene and casts a shadow on the paving")

The model places an added object in perspective and gives it a shadow. Say exactly where it should go — otherwise the model will choose the spot itself.

### 7. Change the photo style

```
Redraw the whole photo as a hand-drawn anime illustration with soft watercolor
shading. Keep the same composition, the man's recognizable features, beard and
clothes, the cup, the green awning and the sign «COFFEE SHOP».
```

![The same shot as an anime illustration with watercolor shading; the sign «COFFEE SHOP» is preserved](https://ai2face.com/_astro/style-en.9NYvG2mZ_d7zRV.jpg "Anime watercolor: the composition and sign are intact, the face is still recognizable")

For stylized portraits and profile pictures, see our separate guide to [AI avatars](/blog/avatarka-nejroset).

### 8. Replace the text in a photo

```
Change the sign text on the green awning from «COFFEE SHOP» to «TEA HOUSE». Keep the
same font, white color, letter size and position. Change nothing else in the photo.
```

This is the hardest edit: the model has to read the sign, draw a new word in the same lettering and leave the rest of the frame untouched. That's why we ran it through four models — the results are in the next section.

## Which AI changes text in a photo best: 4 models tested

One prompt, one source, four models. Below is the sign from each result at 100% scale.

![The sign «COFFEE SHOP» and four versions of «TEA HOUSE» from Nano Banana Pro, Nano Banana 2, GPT Image 2.5 and Seedream 5.0 Pro](https://ai2face.com/_astro/text-compare-en.DKdraXi9_8E16p.jpg "Top: the source; below: the models' results. None of them got a letter wrong")

| Model | Text without errors | Font | Position | Rest of the frame |
|---|---|---|---|---|
| [Nano Banana Pro](/models/nano-banana-pro) | yes | slightly thinner than the original | same | unchanged |
| [Nano Banana 2](/models/nano-banana-2) | yes | closest to the original | same | unchanged |
| [GPT Image 2.5](/models/gpt-image) | yes | bolder and narrower | slightly lower | unchanged, but the whole frame came out at a lower resolution |
| [Seedream 5.0 Pro](/models/seedream) | yes | same weight | same | the whole frame slightly shifted and scaled |

The verdict: all four models replace text without errors, in English and, in the Russian version of the test, in Cyrillic. Where they differ is the typeface, and that varies from run to run: here Nano Banana 2 matched the original font most closely and [GPT Image 2.5](/models/gpt-image-2-5) made the letters bolder, while in the Russian run GPT Image 2.5 was the closest. If the sign is part of a brand identity and the font matters, try two or three models on your sign. If you have lots of edits and price matters, Nano Banana 2 is enough.

## How to write a prompt that edits a photo without ruining it

A three-part formula works:

**what to change + what the result should look like + what to keep exactly as is**

- **What to change.** Name the object by its place in the shot: “the passer-by on the left,” “the text on the awning,” “the blazer.”
- **What it should become.** Color, material, time of day, weather: “a cream knitted sweater,” “a warm evening with the street lamps on.”
- **What to keep.** The most important line of the prompt: “keep the face, expression, pose and composition exactly as in the photo.” Without it the model may “improve” the person.
- **One edit per request.** If you need to change both the background and the outfit, do it step by step: send the result of the first edit into the second. That way it's easy to see which instruction didn't work.
- **Any language works.** The models in our test understand prompts in many languages; put text that should appear in the image in quotation marks.

If the photo is blurry or old, that's a different task — see our guides to the [AI photo enhancer](/blog/uluchshit-kachestvo-foto-nejroset) and [AI photo restoration](/blog/restavraciya-foto-nejroset). To try a new haircut or hair color, there's a separate walkthrough: [AI hairstyle try-on](/blog/podobrat-prichesku-nejroset).

## Which AI to choose for photo editing

| Task | Model |
|---|---|
| Edits with a person in the shot: background, outfit, light | [Nano Banana Pro](/models/nano-banana-pro) |
| Lots of simple edits: remove an object, change the background | [Nano Banana 2](/models/nano-banana-2) — cheaper and faster |
| Signage where the original font matters | GPT Image 2.5 or Nano Banana 2 — compare on your sign |
| Edits based on several references, up to 10 images | [Seedream 5.0 Pro](/models/seedream) |
| A product photo on a new background, listing images | Nano Banana Pro; more in our guide to [AI product cards](/blog/kartochki-tovarov-nejroset) |

## Limitations

- **The AI redraws the whole frame.** Even with “change nothing else,” small details — folds, reflections, foliage — may differ slightly from the source.
- **The font is only roughly preserved.** The letters will be correct, but the typeface may change, as it did with GPT Image 2.5 in the English run and with the [Nano Banana](/models/nano-banana) models in the Russian one.
- **Complex objects in hands.** Fingers, jewelry and small text on clothing are the first candidates for artifacts. Check them at full size.
- **Only your own photos, or with consent.** Changing another person's appearance without their permission isn't editing — it's forgery.

## How to edit a photo in AI2Face

Nano Banana Pro, Nano Banana 2, GPT Image 2.5 and [Seedream 5.0](/models/seedream-5-0) are all in one AI2Face generator:

1. Upload the photo as a reference.
2. Choose a model and a resolution — up to 4K.
3. Paste a prompt following the “what to change + what it should become + what to keep” formula.
4. If you like the result, send it into the next edit straight from the feed.

The cost of each edit is shown before you start, you can pay by card or crypto, and credits from packs don't expire. After generation, AI2Face strips AI metadata from the file, so social networks won't find an AI label in it.

## Bottom line

An AI photo editor changes a photo from a single sentence, and in our test all eight edits worked on the first try: background, object removal, outfit, light, season, a new object, style and text. The face and pose never changed — thanks to the “keep it exactly as in the photo” line. For edits with people, use Nano Banana Pro; for a batch of simple edits, Nano Banana 2; for signage in a brand font, GPT Image 2.5. Start with one edit on your own photo and add the next ones step by step.

## FAQ

### How do I edit a photo with AI?

Upload the photo to the generator, pick an editing model such as Nano Banana Pro and describe what to change and what to keep: “replace the background with a seaside promenade, don't change the face, pose or clothes.” You get a new version in 20–40 seconds.

### Can I edit photos with AI for free?

Partly. AI2Face gives you welcome credits after signing up — enough for several edits in Nano Banana Pro or Nano Banana 2. The cost of each edit is shown before you start.

### Will the AI change the face in my photo?

It can, unless you tell it not to. Add “keep the face, expression and pose exactly as in the photo” to the prompt. With that line, the face didn't change in any of our eight edits.

### Can AI change the text in a photo?

Yes. All four models in our test — Nano Banana Pro, Nano Banana 2, GPT Image 2.5 and Seedream 5.0 Pro — replaced the sign «COFFEE SHOP» with «TEA HOUSE» without a single typo, and in the Russian version of the test they did the same with Cyrillic. In the English run Nano Banana 2 stayed closest to the original font, in the Russian run GPT Image 2.5.

### Which AI is best for editing photos?

For edits with people in the shot, Nano Banana Pro: it holds the face and lighting most accurately. For a batch of simple edits, Nano Banana 2 — it's cheaper. For signage in a brand font, GPT Image 2.5.

## Sources

- https://ai.google.dev/gemini-api/docs/image-generation
- https://blog.google/innovation-and-ai/products/nano-banana-pro/
- https://developers.openai.com/api/docs/guides/image-generation
- https://seed.bytedance.com/en/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro