A flat lay tells a shopper what a garment is. An on-model image tells them what it does: where the hem falls, how the sleeve sits, whether the fabric hangs or holds. That second image used to need a model, a studio and a day; it now needs the flat lay you already have and a decision about who wears it.
This page is a method for putting clothing on AI models from a flat lay or mannequin photograph. It publishes no figure for what on-model imagery does to conversion. It describes what the source photograph has to contain, how the generation is set up, and what to check before the result goes on a product page.
How do you put clothing on an AI model?
You photograph the garment flat or on a mannequin, front and ideally back, on a plain background at the camera's full resolution, and upload it to an AI clothing studio. You choose the garment type, choose or build the wearer, set the fit and the background, and generate. The tool keeps the garment as photographed and supplies the body, the pose and the scene. Every output is then inspected against the source photograph at full size for the things that go wrong when a garment is dressed onto a body: neckline, length, print, closures and colour.
What is flat-lay-to-model generation?
Flat-lay-to-model generation is the production of an on-model garment image from a photograph of the garment alone — laid flat, hung, or on a mannequin — using an image model that keeps the garment's shape, fabric, colour and details while it renders a wearer, a pose and a background around it. It replaces the model session, not the product photograph: the flat lay remains the record of what the garment is, and every generated image is checked against it.
What the source photograph has to show
The generated image can only be as faithful as the flat lay it comes from.
- The whole garment, in frame, at full resolution. A cropped hem or a sleeve out of frame is a part the model has to invent. Upload at the camera's original size; Photta accepts files up to 10 MB and converts HEIC and WebP on the way in.
- Front and back. The apparel studio can shoot a back view as a second variation for each model, but only from a back photograph; without one the back is a guess. Two flat lays, one upload each.
- Flat, square to the camera, evenly lit. A garment shot at an angle is foreshortened, and the model will dress a body in the foreshortened version. Soft, even light from above; no hard shadows that read as folds.
- The real colour. Fix the camera's white balance and keep a white card in the frame. A garment photographed warm will be dressed warm.
- Closures and hardware visible. Buttons, zips, drawstrings and labels are the details most often redrawn; if the source shows them clearly, the output has no reason to change them.
- A plain background. Pattern or texture behind the garment gives the model something to confuse with the fabric.
From a flat lay: what the flat lay itself must show
A flat lay is the source most brands already have, and the source that hides the most. Laid on a table, a garment shows its front, its print and its colour and hides its volume, its back, its drape and how its closures sit on a body; everything hidden is something the generation has to supply, and what it supplies it infers.
So a flat lay for on-model generation is shot for the tool, not for the feed:
- Fully open and squared. Sleeves out, hem straight, collar as it is worn, nothing folded under, so the garment's outline is its real outline.
- Closures as sold. Buttons done up if the garment is shown closed, a zip visible, so the tool does not decide.
- The back too. A second flat lay of the back, because a back view generated from a front is a back the tool invented.
- No overlap. A scarf laid across a jacket becomes part of the jacket.
- Plain, contrasting ground, even light, a grey card, full resolution — the same rules as any source, because the cut is only as good as the edge.
- A note of the fit. Loose garments photographed flat look fitted once dressed unless the intended drape is known; keep the size chart beside the source and check the generated fit against it, not against the flat lay.
The check afterwards is the garment's, not the model's: outline, colour against the card, print alignment, closure count and position, all compared to the flat lay at full size.
Setting up the generation
In Photta's apparel studio the steps run from the garment type to the piece, the wearer, the shape, the background and the extras, then the details and a review.
- Garment type first. Top, bottom, top-and-bottom set, or one-piece. The type tells the tool where the garment sits on the body and what it should not invent below or above it.
- The piece. The flat lay, and the back photograph if you have one.
- The wearer. Pick a model from the library or build one once with the model builder and reuse it across the range; the same wearer on every product is what makes a collection page read as one brand. The builder costs credits per model and the model is yours to reuse.
- The shape. How the garment fits the body: loose, regular, fitted. This is a decision about the garment, not the model, and it should match how the garment actually fits.
- The background. White is the default, because a product page wants white; choose a described backdrop only for the secondary images.
- The extras. The back shot, if you uploaded the back.
- Details and review. Resolution and the bill — 5 credits per generated image at 2K and 6 at 4K (checked 4 September 2026) — then generate. Several models can be selected in one run, each rendered as its own photograph.
Source, dress, inspect
The loop the checklist below serves, named so the other apparel pages can point at it.
- Source — a flat lay that shows the whole garment, both sides, in its real colour, with closures visible.
- Dress — one wearer, one shape, one background per image; the same wearer across the range.
- Inspect — every output against the flat lay at full size, with the checklist, and regenerate rather than retouch.
On-model image checklist
| Check at full size | Pass | Fail — regenerate |
|---|---|---|
| Garment identity | The same garment: cut, panels, seams | A similar garment; a seam added or removed |
| Neckline and collar | The source's shape, sitting where that neckline sits | Reshaped, deepened, or a collar invented |
| Length and hem | The hem reaches where the garment's length implies on that body | Shortened or lengthened to suit the pose |
| Sleeves | Same length and cuff; folds follow the arm | Cuff redrawn; a sleeve length that changed |
| Print, pattern and logo | Where it is on the flat lay, at that scale, unbroken across seams | Print drifted, resized, mirrored, or dissolved at a fold |
| Closures and hardware | Same buttons, zip, drawstring; same count and position | Buttons added, lost or moved; a zip that became a seam |
| Colour | The source's colour under the scene's light | Shifted hue or a different saturation |
| Fabric behaviour | Hangs, drapes or holds the way that fabric does | A knit that hangs like silk; a denim that drapes |
| Body and hands | Natural proportions, hands with the right count of fingers | Distorted limbs; hands hidden to avoid drawing them |
| Back view | The back photograph, dressed | The front mirrored |
A wrong button count is arithmetic, not opinion. Count on the flat lay, count on the output, and reject on a mismatch.
One shirt, four images
Take a button-front shirt. Photograph it buttoned and flat, front and back, on a pale surface, a white card at the edge of the frame. In the apparel studio choose top, upload both photographs, pick the wearer the range uses, set the fit to regular, keep white, enable the back shot, generate at 4K.
Inspect the front image first: count the buttons, follow the placket, check the collar shape and the cuff. Then the back image against the back photograph: the yoke and any back pleat. Then generate the lifestyle image on a described backdrop from the same setup and compare it to the white-background image rather than to the flat lay, because only the scene should have changed. If the shirt passes on white and fails on the backdrop, the backdrop did it; if it fails on white, the source did.
Four images from two photographs, inspected in an order that tells you what to fix.
Frequently asked questions
Can AI put clothes on a model from a flat lay photo?
Yes. A flat lay photographed square to the camera, at full resolution and in its real colour is enough for an AI clothing studio to render the garment on a chosen wearer with a chosen fit and background. The output is checked against the flat lay for neckline, length, print, closures and colour, because those are the things a garment loses when it is dressed onto a body it was never photographed on.
Do I need a mannequin photo or is a flat lay enough?
Either works as the source. A mannequin photograph already shows the garment with volume, which helps with structured pieces such as jackets; a flat lay is faster to produce and fine for tops, knitwear and most bottoms. Whichever you use, photograph both sides: the back view in the output can only come from a back photograph.
How do I keep the same model across all my products?
Choose one wearer per range and reuse it. In Photta you can build a model once with the model builder and select it for every garment, so the collection page shows one person in many pieces rather than many people. Consistency across a catalogue is its own subject, covered in the guide to generating consistent AI models across a fashion catalogue.
Why does the AI change my garment's details?
Because it fills in what the source does not show clearly. A cropped hem, a button in shadow or a print at an angle is a guess, and the model guesses toward what garments usually look like. Photograph the whole garment flat, square and evenly lit, and check closures, print and neckline on every output at full size.
Should I retouch the generated image if a detail is wrong?
No. Retouching a redrawn button or a shifted print means the published image is two fabrications, the model's and yours. Regenerate from the same source with a narrower change, and keep retouching for the background.
What this page does not claim
That on-model images raise conversion or lower returns; they plausibly do, and this page has no measurement of it. What it claims is that the flat lay decides most of what the model can keep, that one wearer across a range is a decision made once, and that a garment which changed on the way to the body is a rejected image.
To dress a flat lay on a model, see flat lay to model; to build the wearer your range will keep, see the AI fashion model generator.