Design your own photo collage with AI
Choose the pictures. Give them room. Make the story yours.
Yes, you can upload several photographs to ChatGPT and ask it to help you make a collage. Tell it how many photographs you have, the shape of the finished print and what matters in the pictures. Ask for a layout first. You will get a much more useful starting point than “make me a nice collage”.
A good collage lets you enjoy the photographs. You can see the people, a favourite picture gets enough room and the smaller moments have a place around it. Here we show you how to brief AI, check its suggestions, change the layout, add words and keep your real photographs intact when that matters most.
You can also start in our free collage creator, arrange your pictures and download a JPG without buying a print. AI is useful for ideas; a layout made with the original photo layers gives you control over exactly which photographs go on your wall.
Start with the photographs, not a shape you have to force them into.
Pick a day, a holiday, a person or a shared theme. A family story might include a favourite portrait, a place that matters and a small detail that brings the memory back. You can mix occasions, but choose the pictures because they mean something. You do not need to include a weaker near-duplicate just to fill a box.
Use the original files where you can. A main portrait deserves a better starting point than a screenshot from a group chat. Keep a separate folder of untouched originals and work on copies. Our guide to sending full-resolution photographs explains how to avoid losing quality before you begin.
The seven supplied originals used in this guide include full-length pictures, closer faces, different poses and a fabric detail. The blue sky, grass and two patterned outfits tie the series together. They are tutorial photographs, not a suggestion that your family pictures need to match a professional photo shoot.
Separate the photographs you must include from those you could leave out. If one picture matters most, say so. If every person needs equal space, say that instead. This simple decision helps you choose between a main-image layout and a regular grid.
Choose a set that works together
Attach your original photographs, give them visible IDs such as Photo 01 and check the inventory before choosing. This asks for advice, not an image edit.
Upload multiple photos and check the count before designing.
Give your pictures short IDs such as Photo 01, Photo 02 and Photo 03. Keep a note matching each ID to its original file and a short description: “Photo 05 — three people beside the blue car”. Renaming files can help you organise them, but do not assume an AI chat can reliably identify every picture from its filename or read its original dimensions.
Attach the originals using ChatGPT’s attachment control. Upload limits vary with your account, the files and the interface, so there is no fixed number we can promise will work for everyone. If you cannot attach the whole set together, send batches and write: “More photos are coming. Wait until I say all photos uploaded before planning.” Finish with the total and those exact words.
Then ask for a written inventory: one row per ID, a brief visible description and the total received. Check it yourself. Two similar portraits are easy to confuse, and a confident count can still be wrong. If necessary, make a contact sheet with a visible number beside each thumbnail and supply the same ID list in your message.
A contact sheet is excellent for discussing order and balance. It is not a substitute for the individual full-resolution files when building the print. Keep labels outside the photo areas, and keep the originals available for the final assembly. Correct any missing or uncertain ID before asking for a design.
Count and identify every photo before designing
Use this before or during a multiple-file upload. A visible numbered contact sheet can support the inventory, but is not your print source.
Explain both the photo count and the finished shape.
“A heart collage” can mean three different things: a rectangular print with heart-shaped photo cells, lots of ordinary cells arranged into a heart outline, or a single large heart containing the photographs. Tell AI which you mean. The outside proportions of the print and the shape of each photo opening are separate choices.
Give the intended width and height, or at least say portrait, landscape or square. Then explain the photo cells: rectangular, circular, heart-shaped or a mix. A portrait A-series print has a different outline from a square. A circle or heart mask also hides parts of a rectangular photograph, even if the file itself is still a rectangle.
Include the mood, main picture, exact words, background, spacing and any faces or groups that must stay whole. “Eleven photos, each used once, a relaxed portrait arrangement, white background, one message space, no overlapping faces” is a useful brief. Keep important details away from the outer trim and any canvas wrap; confirm the actual product guides before deciding that allowance.
Ask for a layout before generating a picture
State the exact photo count, the outside print shape and the separate shape of the photo cells. Ask for a map before a generated picture.
Balanced and random are not opposites.
A looser collage can look lovely without lining every edge up. The pictures still need a relationship: perhaps one main portrait, a few medium pictures and smaller details that lead your eye around the page. “Random” is more useful when it means varied sizes and an informal rhythm than when it means pictures scattered anywhere.
- Give the eye a starting point. Use one larger picture when a particular moment leads the story. Choose equal cells when everyone deserves similar emphasis.
- Match the opening to the subject. Tall portraits usually suit taller cells. Wide groups need enough width for the people at both ends.
- Make the gaps deliberate. Consistent gutters suit a grid. A looser composition can vary them, but avoid isolated pictures drifting in large, unexplained gaps.
- Check the crop, not just the frame. Keep complete heads and meaningful hands visible. Do not move a group so one person disappears behind a heart’s edge.
- Choose useful picture sizes. Tiny faces may disappear on a small print, however many photographs the layout technically holds.
White space gives pictures and wording room. It becomes a problem when it dominates by accident while the photographs are too small to enjoy. Filling every inch is not the goal either. Look at the whole print, then zoom in and check each opening.
The same four photographs, arranged two ways.
We made the examples below using the same four supplied photographs. The first scatters small heart cells across the page and deliberately uses poor crops. The second gives those pictures larger, aligned portrait spaces. These are our own controlled layout examples, assembled from the originals, rather than outputs claimed to come from an AI collage tool.
In the first, one heart cuts into the top of a head and upper face; another clips part of a face at the side. Large uneven gaps take attention away from the photographs. The second keeps all four complete heads and faces, with consistent margins and gutters. No faces were regenerated to make the comparison.
Hearts are still a perfectly good choice. Give the people enough room within the shapes and move each photograph until its crop works. If the mask cannot hold a group comfortably, change the cell or choose another photograph rather than accepting half a person.
A practical ChatGPT workflow: plan first, visual second.
- Upload and identify the originals. Finish the inventory and confirm the total. Tell it which pictures are required and which are optional.
- Set the brief. Give the outside shape, number of photo cells, any message area and your crop priorities. Ask for two layout ideas if you are undecided.
- Ask for a placement map. Give each cell a label, such as R1C1 for row 1, column 1, or A–L for an irregular arrangement. Ask for the photo ID assigned to each cell and any recommended crop.
- Audit the plan. Count the unique photo IDs, list missing IDs and check duplicates. A message space is a message space, not another photograph.
- Request a visual if it will help. Ask for the approved composition as a flat design rather than a room mockup. You can request a blank numbered layout or a generated concept using the photographs.
- Compare and choose your final route. Inspect the result against the originals. Use original layers for the final file when the actual photographs must remain faithful.
Image generation can reinterpret faces, clothing, objects and small details while making a composition. Asking it to keep everything unchanged is sensible, but that instruction is not proof that it did. A blank layout concept avoids this particular problem: use it to decide the arrangement, then place the real files in a collage app or layered editor.
You can also request a bespoke generated visual. Describe actual decisions, such as “one large portrait on the left, three smaller details on the right and a title below”, rather than relying on a style name. Treat the returned picture as a proposal until you have checked both its layout and its photographic content.
Develop a bespoke collage idea
Use this after approving a placement map. A generated visual is a proposal; keep the originals for comparison or faithful final assembly.
Our worked example uses five separate photo uploads.
For the AI composition example at the top of this page, we supplied five original photograph files: one main picture for the left and four supporting pictures for a two-by-two arrangement on the right, with “COLOUR & LIGHT” in a band below. This tests a multiple-photo composition request, rather than asking AI to add words to an already finished collage. Compare the returned result with those originals as well as the brief.
The complete actual result is shown at the top of this page.
The first returned collage included all five requested source scenes once, in the requested positions, and rendered COLOUR & LIGHT correctly. The four portrait heads remained whole; the close fabric detail stayed a detail. The images were scaled and reframed, and fine facial, clothing and background details were generatively reinterpreted. The actual saved output is 1254 × 1254 pixels, so it is a composition example rather than a promise of unchanged photographs or large-print resolution.
The seven originals remain available in the source gallery for the other examples. The eleven-, twelve- and eighteen-photo prompts below are adaptable layout plans, not a claim that we generated a finished collage from eighteen supplied photographs.
Eleven photos in a twelve-space layout: choose the spare space.
If you have eleven photographs but like a layout with twelve spaces, you have a useful choice. Use Photos 01–11 once each and make the twelfth space a message, a date or an intentional blank. Tell AI which you want. It should not quietly repeat a favourite picture or invent a twelfth photograph to make the numbers fit.
A relaxed twelve-space composition can combine larger and smaller cells, provided every ID has a home and the gaps still feel considered. Ask for a map with eleven photo assignments and one cell marked MESSAGE or BLANK. The diagram below explains that counting idea; it is a layout plan, not a claim that we uploaded eleven different photographs for this example.
Use eleven photos in a twelve-space composition
Eleven unique photographs plus one chosen message or blank space. Edit the spare-space choice before sending.
If you actually have twelve photographs, use Photos 01–12 once each. Decide separately whether words should sit outside the twelve-photo arrangement. Otherwise a request for twelve photos “including a message box” can accidentally leave room for only eleven. Our twelve-photo prompt makes the distinction explicit.
Give twelve photographs twelve photo cells
Use this when you really have twelve images. A caption outside the grid does not reduce the photo count.
Eighteen photos in a grid, with the message underneath.
For eighteen photographs, a regular grid makes the count straightforward. Six columns by three rows gives eighteen photo cells for a landscape design. Three columns by six rows gives eighteen for a portrait design. Put the caption in a separate band beneath the whole grid, so it does not replace one of the photographs.
Ask for a row-by-row map. In a simple sequence, the first row contains Photos 01–06, the second 07–12 and the third 13–18. You can then rearrange for colour, chronology or people, but the audit still needs eighteen different IDs, each appearing once. The plan below shows numbered spaces; we have not invented eighteen source photographs to fill it.
Six-by-three describes the grid, not the exact shape of every opening. Those cells may need to be taller or wider depending on the print proportions, margins and caption. Give group photographs wide enough spaces, or choose a mixed layout if a regular grid would crop people badly. More photographs usually need a larger print to give each face useful room.
Plan eighteen photo cells with a caption underneath
Choose landscape 6×3 or portrait 3×6. The message band sits outside all eighteen photo cells.
Our free collage creator gives you a practical starting point.
In our collage creator, choose a layout, add your photographs and adjust the individual crops. Work through the spacing, border and background choices, then check every cell. The first automatic crop is a starting point; you know which people and details need to stay in view.
The seven supplied completed layouts throughout this guide show different ways to arrange the same portrait series. Four equal cells give the pictures similar emphasis. Story Five gives one a larger place. Feature Eight brings in smaller details alongside the portraits. A busier layout is useful when the extra pictures add something to the story.
We have also kept two versions of Story Five. The layout stays the same but the main photograph sits differently. In one, the person is much smaller beneath a large area of sky. Moving the crop brings the subject forward. You can change the feel of a collage without changing the photograph itself.
Our free collage walkthrough explains the app. You can download a JPG without signing up or buying a print; canvas and poster orders are optional paid products. Keep the original photographs as well as the download. A JPG is a flattened picture, not separate editable photo cells or a promise of the same file as a paid production master.
Use a reference to explain a look, then adapt it to your photos.
A reference can show the spacing you like, the balance between one main picture and smaller images, or where a title could sit. Identify it as a reference, separate from the photographs to include. Ask for a new arrangement suited to your own images, without copying another person’s photographs, wording or branding.
A layout full of landscapes may need changing when your files are portraits. All seven originals in our supplied series are portrait photographs; a wide-looking finished cell is a crop. Ask how the proposed design will handle orientation before accepting a visual that looks attractive at thumbnail size but loses someone at the edge.
Use a reference for spacing and composition
Identify the reference separately. Use it to discuss spacing and emphasis, then adapt the layout to your actual photographs.
Change the layout without losing track of the pictures.
Save an approved first version before making changes. Then request one clear revision at a time: “Swap Photos 05 and 12; keep all other assignments”, or “Make Photo 03 larger and move the caption below the grid”. Ask for the updated map as well as a visual, so you can check that nothing went missing during the rearrangement.
Distinguish moving a whole cell from moving the photograph inside it. The first changes the composition; the second changes the crop. “Move Photo 05 down inside its current cell to show the whole head” is more specific than “fix the photo”. If you want a different frame shape, expect to review the crop again.
With a generated image, even a narrow revision can affect other areas. Compare all the faces after each edit, not just the part you requested. In a layered project, select the relevant original-photo layer or text layer and leave the others alone. That makes repeated changes much easier to control.
If one photograph needs work, edit a copy of that individual original before replacing its collage layer. This is easier to inspect than asking AI to repair a tiny panel inside the whole collage. An AI edit can still reinterpret a face even when you request preservation. Our guides to removing a distracting background and improving a photo with AI explain what to request and what to check.
Revise the order or crops without changing everything
Save your approved version first. Name the exact photo IDs and say whether you mean the cell position or the crop inside it.
Add a caption to the free JPG, or keep the text editable.
A name, date or short phrase can finish a collage well. Upload the free JPG to ChatGPT, give the exact wording and say where it should go. A plain caption band below the photographs gives the words room without covering faces. Specify spelling, line breaks, colour and position, and ask for a separate result.
Ask AI to add a caption to your downloaded collage
Upload the untouched JPG download and give the exact wording. An AI text edit still needs a full-photo comparison afterwards.
For the actual edit above, we asked for “COLOUR & LIGHT” beneath a six-photo collage. The wording is correct, but the AI also reframed the grid and changed details in the photographs. The supplied collage is 2048 × 2048 pixels; the returned result is 1254 × 1254. It is a useful design idea, but it is not the original collage with only a text overlay.
To keep the pictures intact, open the downloaded JPG in Photopea and protect its base layer. Add the wording with the Type tool on a new layer. If there is no clear space, use Image → Canvas Size to add a band below the pictures, checking the anchor and extra height. This adds space rather than stretching the photographs.
Save a PSD working file so the new text remains editable, then export a separate PNG or high-quality JPG. Later, reopen the PSD to move, resize or correct the caption. Exported JPG or PNG lettering is flattened into pixels. Our guide to adding words to a photograph covers this further.
Move or correct the caption and keep it editable
This asks for a practical text-layer workflow. Keep your PSD so you can change the words later.
Build an AI-planned collage with the original photos in Photopea.
If the photographs themselves must stay faithful, use AI to plan and assemble the real files separately. You can follow a supported layout in our app, or build a custom arrangement in Photopea. A flattened collage JPG cannot be reopened as individual editable cells; start from the original photographs for that level of control.
- Create the document. Choose File → New and enter the agreed pixel width and height. Set the background and reserve the margins, gutters and caption area before filling the page.
- Make the cell shapes. Use the Rectangle tool in Shape mode for each opening, with guides for aligned edges. Name the shapes by your placement map. Use a circle or heart only where its crop suits the photograph.
- Place an original as its own layer. Use File → Open and Place, or drag the photo into the open document. Name its layer with the photo ID. Keep one photograph per layer so you can find and change it later.
- Clip and position it. Put the photo directly above its cell shape and enable Layer → Clipping Mask. Use Edit → Free Transform with the proportions linked to resize uniformly, then move it to show the important details. Confirm the transform and check the mask boundary.
- Repeat from the map. Work cell by cell, checking groups and faces. Group each cell with its photo layer if helpful. Add a separate Type layer for the caption.
- Keep the working file. Save as PSD, then export a separate print image. Check the export dimensions rather than assuming they match your project.
Clipping hides parts outside the opening without requiring AI to redraw the person. Use normal proportions and avoid generative tools if you want a faithful assembly. Resizing and export still resample pixels, so “original layers” means preserving the photographic content, not that every output pixel will be identical to the camera file.
Check both the full print and the resolution inside each cell.
Open your preferred ChatGPT image and use Save to download the actual file rather than taking a screenshot. Keep versions such as collage-v01, collage-v02 and collage-final alongside the originals and any PSD. Check the saved width and height after each AI edit, crop or export. Asking for “4K” or “300 DPI” does not establish the actual pixel dimensions you received.
The full canvas needs enough pixels, but each cell needs useful detail too. A huge document cannot restore detail missing from a tiny inserted photo. The relevant number is the source pixels left after cropping compared with that cell’s printed size. A heavily zoomed crop may use only a small portion of an otherwise large original.
The eighteen-slot diagram uses an A3 landscape planning document: 420 × 297 mm, or about 4961 × 3508 pixels at 300 pixels per inch. Its 150-pixel outer margins leave a 4661 × 2628-pixel photo area above a separate caption band. With 40-pixel gutters, each of the six-by-three cells is about 744 × 849 pixels. At this planned print size, that is roughly 63 × 72 mm per cell. Check how many useful source pixels remain after each crop; these planning figures are not a universal quality threshold.
Look at every face, each cell edge and the wording. Count the photo IDs again. Check that a repeated picture is intentional and nothing important has disappeared. Our photo-quality guide and print-size guide and calculator help you judge the file.
Check the complete collage before choosing a print
Attach the proposed file and the originals or labelled contact sheet. This asks for an audit, not another redesign.
If you made the collage in our app, follow its review and print options. An externally finished collage is one picture file: upload it through the ordinary photo-print route. Importing that picture will not turn it back into separate editable collage cells.
Choose a poster for smooth paper or a canvas print for depth and texture. Review the whole front and the side guides before approving a canvas. If you are unsure whether the smaller pictures will print clearly, ask us with the file and intended size.
Collecting pictures from friends? Our birthday collage guide helps organise them. If you would rather make the words the main feature, try making Word Art with AI, or explore our collage canvas page for a starting point.
Questions about making and printing your collage.
Can I upload several photos to ChatGPT and ask it to design a collage?
Yes. Upload your photographs, give them clear IDs and explain the count, outside shape, cell shapes and any text. Ask it to confirm the inventory and propose a placement map before generating a visual. Upload availability and limits vary, so use batches when needed and tell it to wait until the complete set has arrived.
What should I do if I cannot upload all my photos at once?
Send labelled batches and state the expected total. Ask ChatGPT to wait for “all photos uploaded”. Then check a written inventory. A numbered contact sheet can help discuss the whole set, but keep the individual originals for final assembly; tiny contact-sheet thumbnails should not become the source for your printed collage.
Can I use eleven photographs in a twelve-space design?
Yes. Use each of the eleven once and choose a message or intentional blank for the spare space. Label that cell clearly in the plan. If you have twelve photographs, ask for twelve photo cells instead and put any caption outside them. Do not let a missing picture be filled by an unnoticed duplicate.
What is a good layout for eighteen photographs?
A six-column, three-row grid is a clear landscape starting point; three columns by six rows suits portrait. Put the message in its own band outside the eighteen cells. Check the proportions of your photographs and the final print size before committing. Wide groups may need a mixed layout rather than narrow, equal cells.
Does a heart collage have to use heart-shaped photo cells?
No. You can arrange rectangular photographs into an overall heart, clip each picture into a heart, or place several photographs within one large heart. Explain which you mean. Also give the outside print proportions. Every decorative mask needs a crop check, particularly with groups or faces close to the photograph’s edge.
Can I ask for a random layout that still looks balanced?
Yes. Describe varied picture sizes, a clear main photograph, deliberate gaps and no unwanted overlaps. Ask for a cell map so every photo has a place. “Random” alone gives little guidance about the important things: which people need room, where your eye should begin and how the pictures fit the finished shape.
Will AI keep the faces and photographs exactly the same?
Do not assume so. A generated visual may reinterpret details, and further edits can affect areas you did not ask to change. Compare the result with the originals. When faithful photographs are essential, let AI help with planning and assemble the original files as separate layers or in supported collage-app cells.
How do I check that no photos are missing or repeated?
Use a written ID-to-cell map and count the unique IDs. For eighteen photographs used once, there should be eighteen photo assignments and eighteen unique IDs. A message band does not count as a photo. Compare each cell visually with its original as well; matching numbers in a plan do not prove a generated picture followed them.
Can I change just one picture or move a caption later?
In a layered project, select that photo or text layer and make the change there. In ChatGPT, identify the exact ID or caption and ask for a narrow revision, then inspect the whole result. Save versions so you can return to a better crop or layout if a later edit changes more than intended.
Can I download a free collage and add editable text?
Our creator offers a free JPG. Open it in Photopea and add a separate Type layer, then save the working file as PSD. The caption remains editable there. A JPG or PNG export is flattened, and the underlying JPG does not contain separate editable photo cells. Keep both your working file and the untouched download.
Does asking AI for 300 DPI make a collage ready to print?
No. Check the actual saved pixel width and height, the intended print size and the detail in each cropped photograph. A resolution label does not create missing detail. The whole file can have many pixels while one small or heavily enlarged source still looks soft. Use the print-size guide or send us the file if unsure.
Can I print a collage made in ChatGPT or another editor?
Yes, you can upload a suitable finished picture file for a canvas or poster. Download the full file, check its dimensions, all faces, the photo count and the wording first. Use the ordinary photo-print route for an externally made collage; uploading a flattened image does not recreate editable cells inside our collage creator.
Found this guide useful?
Add Your Image 2 Canvas as a preferred source on Google.
3 for 2 · Mix and match
Canvas prints.
Quality you can see.
Printed and finished by our family team in West Yorkshire.
From £14.99, including free standard UK delivery.
Create your canvasMix qualifying physical Canvas and Poster prints. Cheapest qualifying base print free; upgrades and extras remain payable. Downloads excluded. Delivery exclusions apply.