The photo that makes AI try-on look real
Almost every disappointing AI try-on is a photo problem, not a model problem. Stand square to the camera in even light, keep your arms away from your sides, wear something close-fitting, and include your whole body — that single image sets the ceiling on everything you generate afterwards.
In Utopia the photo you upload is not used once. It builds your fit model — the reusable representation of your face, proportions and skin tone that every future try-on is anchored to. Get it right once and every garment you ever try looks right. Get it wrong and you will spend weeks wondering why the results are almost, but not quite, convincing.
So it is worth ninety seconds of care. Here is what actually moves the needle, roughly in order of how much difference it makes.
1. Full length, head to feet
This is the one people skip and the one that costs the most. A model that cannot see your legs has to guess your leg length, and it will guess average. Every trouser, every midi dress, every coat hem will then land in a plausible place rather than your place — which defeats the entire purpose, because where a hem lands is most of what you were trying to find out.
Include your feet. Leave a little space above your head. A phone propped against a wall at roughly hip height, about two metres back, gets it in one.
2. Arms away from your sides
Arms pressed flat against your body merge with your torso in the image, and the model has to work out where one ends and the other begins. It usually can — but it does it by inference, and inference is where errors live. A gap of even ten or fifteen centimetres between each wrist and your hip removes the ambiguity entirely.
Do not pose. Crossed arms, hands on hips and hands in pockets all hide the geometry the model needs. Arms relaxed and slightly out, like you are standing in a lift and mildly bored, is the ideal.
3. Square to the camera
A three-quarter turn is more flattering and much harder to work with. Turned away from the lens, one shoulder is closer than the other, one hip is hidden, and the model has to reconstruct a side of you it cannot see. Face the camera straight on, shoulders level, weight even on both feet.
The camera should be at roughly hip to waist height and level — not tilted up or down. A phone held at eye level and angled downwards foreshortens your legs, and the fit model inherits that distortion permanently.
4. Flat, even light
Every generated try-on has to reconcile the garment's lighting with your photo's lighting. The flatter and more even your original, the more room the model has to relight the scene coherently. Hard directional light bakes a shadow into your fit model that then has to be argued with in every image afterwards.
- Best: an overcast day outdoors, or indoors a couple of metres back from a large window, facing it.
- Fine: a bright room with the ceiling light on and no single dominant source.
- Avoid: direct sun (hard shadows across the face and body), a window directly behind you (you become a silhouette), and a single lamp to one side.
- Avoid: phone flash. It flattens skin tone and produces a hard shadow outline behind you that the model reads as part of your shape.
5. Wear something close-fitting
The model has to infer your body shape through whatever you are wearing. A baggy hoodie hides your shoulder line and your waist, and the fit model that results will be built around a guess. Fitted clothes — a t-shirt and jeans, leggings, activewear — give it your actual silhouette.
You do not need to wear anything revealing. A plain t-shirt is enough. The rule is only that the clothes should follow your shape rather than replace it.
6. Plain background
A blank wall is ideal. A busy background is not fatal, but anything close behind you in a similar colour to your clothes makes the boundary between you and the room ambiguous — and an ambiguous boundary is the thing most likely to produce a soft, slightly melted edge in the output.
Nothing in front of you either: no bag over the shoulder, no coffee cup, no dog. Anything you are holding has to be reasoned about in every subsequent image.
The quick version
| Do | Don't |
|---|---|
| Whole body, feet included | Cropped at the knee |
| Arms clear of your sides | Arms crossed or in pockets |
| Square to the camera, shoulders level | Three-quarter turn |
| Camera level, at hip height | Phone held high, angled down |
| Soft even light, facing a window | Direct sun, backlight or flash |
| Fitted t-shirt and jeans | Oversized coat or baggy hoodie |
| Plain wall behind you | Busy background, or objects in hand |
| Neutral expression, eyes open | Sunglasses, hat, hair over the face |
How to tell your photo was good
Generate one try-on with something simple — a plain crew tee, a straight-leg jean — before you try anything complicated. If the shoulders sit where your shoulders are and the hem lands where you would expect on your own body, the fit model is sound and everything else you throw at it will work.
If it looks subtly off on a plain t-shirt, it will look badly off on a bias-cut dress or a puffer. Retake the photo rather than blaming the garment. It takes a minute and it is the highest-leverage minute in the whole app.
For the underlying reason any of this matters — why a generated image needs geometry it can actually see — the full explanation of how try-on works covers the pipeline end to end.
Take the photo once. Try on anything, forever.
Get the appFrequently asked
What kind of photo works best for AI try-on?
+
A full-length photo taken square to the camera in soft even light, with your arms held slightly away from your sides, close-fitting clothes and a plain background. Include your feet, and keep the camera level at roughly hip height.
Can I use a photo where I'm wearing a coat?
+
It will work, but poorly. Loose or bulky clothing hides your actual silhouette, so the model has to infer your shoulder line and waist rather than see them. A fitted t-shirt gives noticeably better results.
Does the background matter for AI try-on?
+
A plain wall gives the cleanest result. A busy background is usually fine, but anything directly behind you in a similar colour to your clothes blurs the boundary between you and the room, which can produce soft edges in the output.
Do I need a new photo for every try-on?
+
No. Utopia builds a reusable fit model from a single photo, and every try-on afterwards uses it — which is why that one photo is worth taking properly.
See it on you, not on a model.
Utopia puts any of 400,000+ pieces on your own photo in about a minute. Free on the App Store.
Get Utopia