How to Merge Photos with AI: 11 Copy-Paste Prompts and Guide (2026)

Merging photos with AI means turning two or more separate photos into one realistic shot: you bring people photographed at different times and places, a product and a model, or a person and a location into the same scene. The difference from a classic collage is simple. In a collage, two pictures sit next to each other. In a merge, the model recalculates light, shadows, perspective and scale and produces a single photograph. Everything depends on the instruction you write, the prompt.

In this guide you will find 11 ready-to-use prompts. All of them are written to be copied and pasted straight into ChatGPT or Gemini; just replace the fields in square brackets with details from your own photos. We also added the limits we verified in each tool's official documentation, the reasons facial features get distorted, and what to watch for when uploading someone else's photo in Turkey. Data is current as of September 9, 2026.

What Is AI Photo Merging and How Does It Work?

AI photo merging means giving several photos to an image model that accepts multiple image inputs and asking for one output. The model extracts each subject's identity, texture and color from every photo, then generates the scene you described from scratch while trying to keep that information intact. That is why the quality of the result depends on two things: how sharp your uploaded photos are and how clearly your prompt describes the scene.

In practice the model does three jobs at once. First, identity transfer: the face from photo A has to be recognizable as the same person in scene B. Second, light matching: if the two photos were taken in different light, the shadows have to be rebuilt around a single source. Third, scale and perspective: the height ratio of the two people and their distance from the camera have to be believable. When a result looks fake, one of these three has almost always failed, and fixing that item with one sentence in the prompt is usually enough.

How to Merge Photos: 6 Steps

Short answer: upload sharp photos, say who or what is in each photo one by one, add the identity preservation line, describe the scene including the light direction, specify the output format, and improve the result with a single correction sentence. The six steps below are also the logic behind the prompts further down.

1. Pick your photos: use shots where faces are clearly visible, not dark, and ideally taken from a similar angle. No prompt can rescue a very low-resolution face.
2. Pick the tool and model: for realistic merges of people, Nano Banana Pro on the Gemini side gives more consistent results; for jobs that need precise edits, ChatGPT's image editing model does.
3. Label the photos: in the first line of the prompt, write who is in which photo. The model follows the order exactly as you state it.
4. Add the identity preservation line: state clearly that facial features, skin tone, hair and distinguishing details like glasses must be kept and that no beautifying is allowed.
5. Describe the scene and the light: location, pose, the direction the light comes from, how hard the shadows are, and the height ratio of the two people. These four items account for half of how realistic the result looks.
6. Constrain the output: write the aspect ratio, the resolution and what you do not want (text, watermark, frame, collage). If you do not like the result, send a short sentence that fixes the one problem item rather than rewriting from scratch.

11 Ready-to-Use Photo Merge Prompts (Copy and Paste)

Copy the prompts as they are, replace the fields in square brackets with your own details, and send them after uploading your photos. Remember to delete the brackets; the model sometimes treats a square bracket as a label that should appear in the image. The prompts are written in plain language, and adding your own photography terms for light and lenses will not hurt the result.

Prompts for putting two people in one frame

I'm uploading two separate photos. The person in the first photo is [Person A], the person in the second photo is [Person B].

Your task: merge these two people into a single, realistic photo.

Identity preservation: keep both people's facial features, eye and nose proportions, skin tone, hairstyle and distinguishing details such as glasses, beard or moles exactly as they are. Do not beautify, slim, rejuvenate or age the faces.

Scene: [scene description, e.g. standing side by side in a park in autumn].
Pose: the two people are close to each other, their shoulders at the same height, with a natural gap between them.
Light: match the lighting of both photos to a single source; the light comes [from the upper left] and the shadows are soft.
Perspective: both should look as if they were shot with the same camera at the same focal length; their height ratio should be realistic.

Output: a single photo, [16:9] aspect ratio, realistic photo quality. Do not add a collage, frame, text or watermark. When you are done, tell me in one sentence what you took from each photo.

This prompt is the base of the list: the other ten are this skeleton adjusted for different scenarios. If the result looks like a collage, strengthen the perspective line. If the faces do not look familiar, add the sentence 'follow the face in the reference photo pixel by pixel' to the identity preservation line.

Of the two photos I uploaded, the first is [Person A] and the second is [Person B]. They are a couple.

Your task: merge them into one frame as a natural, intimate couple photo.

Identity preservation: keep both faces exactly as they are. Do not change facial proportions, skin color, hair color or hairstyle; do not add makeup.

Scene: [location, e.g. a sunset by the sea].
Pose: [Person A] slightly in front, [Person B] half a step behind; one is touching the other's shoulder. Both are looking at the camera.
Clothing: keep the clothes from both photos, only adapt the wrinkles and shadows to the scene.
Light: [golden hour] light, warm tone, even lighting on the faces, no hard shadows.

Output: a single photo, [4:5] vertical, realistic. Do not add hearts, text, stickers or filters.

I'm uploading [3] separate photos. Each photo shows one member of our family: [photo 1: mother], [photo 2: father], [photo 3: child].

Your task: merge them all into a single family photo.

Identity preservation: keep each person's face as it looks in their own photo. Their age, apparent weight, hair and skin tone must not change. Do not make anyone look like someone else.

Arrangement: place them so that [mother and father are in the middle and the child is in front]; height differences should match real life.
Scene: standing [in front of the sofa in the living room].
Light: a single soft indoor light; every face is lit at the same warmth, and no face is brighter than another.

Output: a single photo, [3:2] horizontal, realistic family portrait. Do not add a studio backdrop, frame or text.

In group photos, face quality drops as the number of people grows. Instead of merging more than three people in one go, first generate pairs and then merge those in a second request. Facial features hold up far better that way.

One frame with people far away or from the past

Of the two photos I uploaded, one is [Person A: me] and the other is [Person B: my sibling who lives abroad]. The two photos were taken in different places at different times.

Your task: create a single frame that looks as if we met in the same place at the same moment.

Identity preservation: keep both faces exactly as they are; you may change the expressions, but do not change the facial features.

Scene: [airport arrivals hall], background slightly blurred.
Pose: [hugging], faces turned toward the camera.
Light and color: if the two photos have different color temperatures, match both to the light of the scene; balance the skin tones as if both people were under the same light.
Realism: the shadows should match the floor; leave no cut-out edges or halos around the people.

Output: a single photo, [16:9], documentary-style realism. Do not add text, a date stamp or a watermark.

I'm uploading two photos. The first is an old, worn photo of [my late grandfather]. The second was taken today and shows [me].

Your task: merge the two people into one frame as a respectful, natural family photo.

First restore the old photo: remove scratches, stains and creases, increase sharpness, correct the yellowing. But do not change the facial features, the expression or the age; do not invent details that are not there.

Identity preservation: both faces stay exactly the same. Do not make the person in the old photo look younger or more modern.

Scene: [a plain indoor space with light-colored walls].
Pose: side by side, shoulder to shoulder, both looking at the camera.
Light: light both faces with the same soft light; adapt the gray tones of the old photo to the colors of the scene, but do not over-colorize.

Output: a single photo, [3:2], realistic. Do not add symbolic elements such as angels, beams of light or text.

Prompt 5 does two jobs: it restores the old photo first, then merges. To stop the model from inventing details during restoration, keep the line 'do not invent details that are not there' in the prompt. Without it, the model adds its own guessed features to low-resolution faces.

For businesses: merging people, products and places

I'm uploading two photos. The first shows [product: our own product], the second shows [model: a member of our team].

Your task: create a single promotional photo that shows the product realistically in the model's hand or on the model.

Product fidelity: keep the product's shape, color, material texture, label and logo exactly as they are. Do not rewrite or change the text on the product.
Identity preservation: keep the model's face and skin tone exactly as they are.

Arrangement: the product is [in the model's right hand], with its front facing the camera and visible; the fingers must not cover the product's logo.
Scene: [plain light gray studio backdrop].
Light: soft key light [from the upper left], a little fill light from the right; no glare on the product, and the shadow falls naturally on the floor.

Output: a single photo, [4:5] vertical, good enough for e-commerce and social media. Do not add price tags, promo text or watermarks.

I'm uploading two photos. The first shows [person], the second shows [place: somewhere I want to go].

Your task: place the person inside this location as if the photo had been taken there.

Identity preservation: do not change the person's face, hair, clothing or body proportions.
Location fidelity: keep the architecture, colors and layout of the place; do not add new buildings, signs or objects.

Arrangement: the person stands [in the foreground, in the left third of the frame]; the place behind them stays clearly visible.
Scale: the person's height is consistent with the size of [a door, railing, person] in the background.
Light: match the light direction on the person to the light of the place; their shadow falls on the ground in the same direction as the other shadows in the scene.
Color: slightly adapt the person's skin tone and clothing colors to the color temperature of the place; do not oversaturate.

Output: a single photo, [16:9], realistic travel photo. Do not add text, location tags or frames.

In product images, the most important line is 'do not rewrite the text on the product'. Image models tend to regenerate label and logo text from their own guesses, and without this line your brand name may come out garbled. For commercial use, zoom in on the label and check it before publishing the output.

Collage and before-and-after comparison

I'm uploading [2] photos: [photo 1: before], [photo 2: after].

Your task: combine these two photos into one image in a side-by-side comparison layout.

Layout: the first photo on the left, the second photo on the right; both the same width, with a thin [8 pixel] white gap between them.
Alignment: align the main subject of both photos at the same height and the same scale; crop the edges if needed, but do not cut off the subject.
Color: bring the brightness and color temperature of the two photos closer together; do not change the content and do not retouch.

Output: a single image, [16:9] horizontal. Do not add text, arrows, logos or watermarks on top. Do not regenerate the content of the photos, only place them.

Not every job needs a single merged shot. For before-and-after posts, portfolios and product variants, a collage is the more honest presentation, and because the model does not regenerate the content, your original photos stay exactly as they were.

Versions by tool: Gemini, ChatGPT and a quick version

Use these two photos as references: the first photo is [Person A], the second photo is [Person B].

Character consistency: generate both people with exactly the facial structure shown in the reference photos. Do not reinterpret or stylize the facial features, and do not give them a cartoon or illustration look.

Composition: the two people sit [on either side of a table in a cafe], placed according to the rule of thirds.
Camera: 50 mm lens look, eye level, f/2.8 depth of field; softly blurred background.
Light: soft side light from a window, warm white balance.
Texture: close to real photo grain, skin texture preserved, no excessive smoothing.

Output: [16:9], [2K] resolution, a single photo. No text, logo or collage.

Use the two photos I uploaded. First describe both: who is in each photo, where does the light come from, what is the camera angle?

Then merge these two people into a single realistic photo.

Identity preservation: keep both faces exactly as they are; age, weight, skin tone, hair and details like glasses must not change.
Scene: [scene description].
Light: take the lighting of whichever photo suits the scene better as the base, match the other one to it, and tell me in writing which one you chose.
Scale: set the height ratio of the two people as it would be in real life.

Output: a single photo, [1536x1024] horizontal, realistic.

After generating the image, also write this: which parts were hard to merge, and which sentence should I add on the second try?

I'm uploading two photos. Merge these two people into one frame: [scene], [light direction], [aspect ratio]. Keep both faces exactly as they are, do not beautify them, and do not add a collage or text.

Prompt 9 takes advantage of the way Gemini handles reference images separately as characters, objects and styles; it responds strongly to camera and lens descriptions. Prompt 10 plays to ChatGPT's strength: have it read and describe the photos first, then merge them. Prompt 11 is a shortened version for quick tests inside a chat.

The 6 Blocks of a Photo Merge Prompt That Works

All the prompts above use the same skeleton. Once you learn it, you can write your own prompt for scenarios that are not on the list. The table shows what each block does and what the model does when you leave that block empty.

ChatGPT, Gemini or CapCut? Which One for Which Job

All three tools merge photos, but each is good at different jobs. On the Gemini side, the Nano Banana family was designed to work with multiple reference images and has a wide range of resolution options. ChatGPT's image model stands out for precise edits, and because it can read and describe the photos first, it can tell you when it has not understood what you want. Ready-made tools like CapCut produce template-based collages for people who do not want to write prompts.

The choice comes down to this: if you really want two people to meet in the same scene, start with the Nano Banana models on the Gemini side; if you need to change a product label, some text or a small area without breaking it, ChatGPT's editing model is safer; if a side-by-side collage is all you need, a ready-made tool with no prompt at all saves time.

Identity Preservation: Why Do Facial Features Get Distorted?

The most common complaint about photo merging is always the same: the result looks great, but those people are not us. The technical reason is that the model regenerates the face. It draws a new face from the features it extracted from the reference photo, and if the prompt sets no limits, that new face drifts toward the average attractive face in its training data. The table below lists the four most common distortions and the sentence that fixes each one in the prompt.

7 Techniques That Improve the Result

• Upload photos without cropping them: when the model sees the context around the face, you get fewer halos and cut-out marks along the edges.
• Fix the light direction in one sentence: a clear direction like 'the light comes from the upper left' makes the two subjects sit in the same scene.
• Give a scale anchor: a reference like 'their height is consistent with the door behind them' noticeably reduces proportion errors.
• Put negative instructions at the end: if you do not want text, a watermark, a frame or a collage, list them on one line in the output block.
• Ask for one fix at a time: when you do not like the result, change only the problem item and leave the rest of the prompt as it is.
• Run the same prompt twice: image models work on probabilities, and a second attempt with the same prompt often gives a better result.
• Zoom in and check the output: hands, glasses frames, ears and product labels are the four places where errors show up most often.

The common logic behind these seven points: wherever you leave the model room to be creative, it applies its own preferences. A good merge prompt tells the model where it has no freedom.

6 Common Photo Merging Mistakes

• Uploading low-resolution or dark photos: if the face data is not there, no prompt can bring it back.
• Skipping the identity preservation line: this is the number one reason results do not look familiar.
• Trying to merge shots taken from very different angles: placing a face photographed in profile into a front-facing scene usually does not work.
• Asking for five things in one prompt: when you say merge, change the background, change the clothes, make them younger and sharpen it, the model gives a mediocre result on all of them.
• Not deleting the square brackets or sending them unfilled: the model may write the label inside the brackets onto the image.
• Publishing output for commercial use without checking it: always zoom in on the logo, label text and hand details before publishing.

Using These Tools from Turkey: Consent, KVKK and SynthID

The legal side is the part of photo merging people skip most often. Under Turkey's Personal Data Protection Law (KVKK), a facial image is data that can directly identify a person, and when it is processed as biometric data it counts as special category personal data. In practice this is simple: before uploading a photo of a friend, employee or customer to an AI tool, you need their consent. For business use, this consent is expected to be in writing and to state the purpose for which the photo will be processed. This is not legal advice; for data controller obligations in business processes, consult your legal advisor.

It is now also technically possible to tell that a generated image is AI output. Google states in its official documentation that it adds a SynthID watermark to all images generated with Gemini. The watermark is invisible to the eye, but it can be detected when you upload the image to Gemini and ask whether it was generated by AI. If you use merged images in advertising or corporate communication, stating that the image was produced with AI assistance is the safer choice, both for trust and with possible transparency regulations in mind.

The SynthID watermark was designed to withstand changes such as cropping, adding filters, changing frame rates or lossy compression.

What decides the result in photo merging is how clear your prompt is. Adapt the 11 prompts above to your own photos, fix the one item you did not like after the first try and send it again, and if someone else's face is involved, get their consent before you upload. If you want to produce product images, corporate portraits or campaign visuals for your business with an AI-assisted workflow, we can set up the whole process and hand it over to your team.