Back to Blog
Tutorials

Virtual Try-On Quality-Control Checklist

A catalogue image is checked against the garment. A try-on is checked against the garment and the person, because the person is the thing the model was told to keep and the first thing it quietly alters. One checklist for every try-on image, in the order that finds the expensive faults first.

Photta TeamEditorial team

12 min read

When a garment is dressed on a catalogue model, the model is the tool's to draw and the garment is yours to check. A try-on reverses half of that. The person is now yours too — a model you photographed and kept, or a shopper who uploaded a photo of themselves — and the tool was told to leave them alone. It does not always. A jaw softens, a shoulder narrows, the hair changes length, the room behind them loses a window. None of it looks wrong. All of it is a different person, and the person will notice.

This page is a checklist for try-on images, the ones where the wearer is fixed. The general checklist for generated garment images is the AI fashion photography quality-control checklist; this one adds what a fixed person changes, and what to do when the images are made on your product pages by shoppers you never see.

How do you quality-check a virtual try-on image?

Put the output beside two sources: the garment's flat lay and the person's original photo. Check the person first — face, body, skin, hair, background and light all unchanged — because a changed person is the most expensive fault and needs no further inspection. Then check that the original clothing is fully gone, then the garment against its flat lay, then the garment on this particular body: length, fit and the edges where it meets skin, hair and hands. Finally check the join: shadow, light and scale agreeing with the photo. Publish on pass; regenerate on any fail; never retouch the person or the garment.

What is quality control for a try-on image?

Quality control for a try-on is the comparison of an output against the garment photograph and the person's photograph, before the image is published on a listing or shown to the shopper who asked for it. It differs from checking a generated catalogue image in what is fixed: there, the garment; here, the garment and the person and the person's photograph. The checks fall into five groups in a fixed order — person, removal, garment, garment-on-body, join — and the outcome is pass, regenerate, or reshoot one of the two sources.

Why the person changes

  • The face is the model's habit. Image models have drawn more faces than anything else, and left to themselves they draw the face they expect, not the one they were given. Small, regular, symmetrical improvements are the tell.
  • The body is reconciled with the garment. Dressing a fixed garment on a fixed body means one of them gives. Sometimes the garment stretches; sometimes the waist narrows to suit the garment.
  • The old clothing leaves traces. A collar that was higher, a sleeve that was longer, a print under a plain new top: what the tool did not fully remove shows through as a seam that should not be there.
  • The background is redrawn around the edit. The area behind the new garment is regenerated, and a doorframe or a shelf that ran behind the old outfit may not survive.
  • The light belongs to the photo. The person was lit by their room. The garment arrives lit by nothing, and the join is where the two lights meet.
A photograph of a person seen from behind pinned to a corkboard beside a fabric swatch and a wooden clothes peg, seen from above.
Two sources, one output: the person and the garment are both fixed.

The inspection order

  1. Person — against the original photo. A changed person is a rejected image, whatever the garment did.
  2. Removal — nothing of the original clothing left.
  3. Garment — against the flat lay, exactly as in the catalogue checklist.
  4. Garment on this body — length, fit and edges for this person's proportions.
  5. Join — shadow, light, scale and the redrawn background.

Virtual try-on quality-control checklist

Person

RowPassReject
FaceThe same face: features, expression, asymmetriesRegularised, smoothed, aged, or a different person
BodySame proportions, shoulders, waist, limbsNarrowed or widened to suit the garment
Skin and hairSame tone, texture, hair length and shapeTone shifted, hair shortened, or hair regrown over the garment
Hands and accessoriesSame hands, rings, watch, glasses, bagRedrawn hands; an accessory lost or added
PoseThe photo's pose, unchangedA limb moved to make the garment easier
BackgroundThe room as photographed, with what ran behind the outfit intactA window, shelf or doorframe lost or bent behind the garment

Removal

RowPassReject
NecklineThe new garment's neckline, and skin or the new garment beneath itThe old collar or strap showing through
Sleeves and hemThe new garment's ends; below them, skin or the original lower garment as photographedA cuff or hem of the old garment past the new one
LayeringOnly the layers that should be thereThe old top's print visible under a new one

Garment

RowPassReject
Identity and detailsThe flat lay's garment: cut, closures, seams, print, colourA similar garment; details moved, lost or added
ColourThe flat lay's colour under the photo's lightShifted beyond what the room's light explains
FabricHangs, drapes and holds as that fabric doesWrong drape; texture smoothed

Garment on this body

RowPassReject
LengthThe hem and sleeves end where that garment's length would on this personShortened or lengthened to a typical point
FitOversized stays oversized; fitted stays fitted, on this bodySilhouette changed to suit the pose or the body
EdgesClean boundaries at the neck, wrists, waist and handsFabric fused with skin; a hand behind fabric that should be in front
OcclusionHair, hands and straps in front of or behind the garment as they were in the photoHair painted over; a bag strap removed

Join

RowPassReject
LightThe garment lit from the same side and with the same softness as the faceA garment lit from elsewhere, or flat-lit on a side-lit person
ShadowThe garment casts and receives shadow as the body doesNo shadow under the collar or arm; a shadow the photo does not have
ScaleThe garment's size in proportion to the personA collar too large for the neck; a print scaled to a different body
Redrawn areaNothing generated beyond the garment's own outlineHalo, blur or invented background around the garment

A changed person fails at row one; count the rest only when the first group passes.

Person, garment, join

The loop the checklist above serves, named so the other try-on pages can point at it.

  1. Person — everything the tool was told to keep, checked against the original photograph before anything else.
  2. Garment — everything the customer will compare to the delivered product, checked against the flat lay, then on this body.
  3. Join — light, shadow, scale and the redrawn edge, where the two sources meet.

Common try-on failures, by garment type

The rows above apply to every garment; the failures cluster by garment type, and knowing which to look for first shortens the inspection.

  • Fitted tops and dresses: the silhouette follows the body the tool imagined rather than the garment's cut — a boxy top made fitted, a shift dress given a waist. Check the outline against the garment's own shape.
  • Loose and oversized garments: volume lost; the drape flattened onto the body. Check that the garment still hangs away from the body where it should.
  • Outerwear: the layer underneath invented — a top the shopper is not wearing, a lining that does not exist — and the closure decided by the tool. Check the closure state and what shows at the neck and hem.
  • Trousers and skirts: the waist moved, the length changed, the hem position on the leg wrong for the stated inseam. Check length against the size chart, not against the image.
  • Prints and stripes: the repeat re-spaced across the body, stripes that bend where the fabric does not. Check alignment at seams and centre front.
  • Knits and textured fabrics: texture smoothed to a flat colour, or a texture invented. Check at full size against the garment photograph.
  • Sheer and layered garments: opacity changed — a sheer overlay made opaque, a lining removed. Check against the garment photographed over a contrasting layer.
  • Footwear and accessories worn with the garment: items added or removed to complete the look. Check that nothing appears that the shopper did not wear or the store did not sell.

Each failure traces to a source: a garment photographed closed, folded, on a busy ground, or without its layers gives the tool something to invent. Fix the source, regenerate, inspect again.

When the images are made by shoppers

A shopper-side try-on on your own product pages — in Photta, Photta Business — produces images you never review before the shopper sees them. Quality control moves upstream and downstream of the generation:

  • Upstream, the product image. The widget dresses the shopper from the product image on the page. A complete, sharp, plain-background garment photograph is the one lever you fully control; fix it before anything else.
  • Upstream, the wearer photo rules. Say what a good photo is at the moment of upload — one person, facing the camera, the body visible where the garment goes, even light — so refusals and poor results are rarer.
  • Downstream, sample the results. Review a sample of try-on results per product type regularly, with this checklist, and record faults by type: "hair regrown over hoods", "hem shortened on maxi dresses". A fault type found in the sample is a product-image fix or a rule fix, applied to every product it can affect.
  • Downstream, read the feedback. If the widget asks the shopper how the result looked, treat that as a signal for which products to sample first, not as the inspection itself.
A folded stack of garments on a shelf with small blank paper tags, and a hand mirror leaning beside them.
Faults are recorded by type and traced to the product image.

One shopper, one dress

A shopper uploads a photo of themselves standing in a hallway, hair over one shoulder, a bag on the other, and tries on a knit midi dress. Person: same face, same shoulders, the hair still over the shoulder, the bag strap still crossing the chest, the hallway's door still behind them. Removal: the collar of their shirt is gone; below the dress's hem their own shoes remain. Garment: the dress's boat neck, its rib, its colour under hallway light. On this body: the hem reaches mid-calf as a midi would on their height; the rib does not cling where the dress is cut loose; the strap sits in front of the knit. Join: lit from the hallway window like the face, a shadow under the arm, no halo. Pass — and had the hair been painted over the strap, or the door behind them straightened away, the fault would have been recorded by type and the product image checked, not the shopper's photograph.

Frequently asked questions

What should I check first in a virtual try-on image?

The person, against their original photograph: face, body, skin, hair, pose and background, all unchanged. A changed person is the most expensive fault and the easiest to miss, because the tool changes people by making them more typical. If the person passes, check that the old clothing is fully removed, then the garment, then the garment on this body, then the join.

Why does the try-on change my face or body?

Because image models reconcile a fixed garment with a fixed person and one of them gives, and because they draw the face they expect rather than the one they were shown. A frontal, evenly lit photograph with the body visible where the garment goes leaves less to reconcile; comparing the output to the photo at full size catches what still changed.

How do I check the try-on results shoppers generate on my site?

You cannot inspect each one before it is shown, so control what you can upstream — the product image and the upload rules — and sample the results downstream by product type, recording faults by type and fixing the product image or the rule that caused them. Shopper feedback, where the widget collects it, tells you which products to sample first.

Should I retouch a try-on image that is almost right?

No, not the person or the garment. A retouched face is a fabrication on the one thing the person will compare to a mirror; a retouched garment is a fabrication on the thing the customer will compare to the delivery. Regenerate from the same two sources, with a narrower request, or fix the source photograph.

How is this different from checking a generated catalogue image?

A catalogue image has one fixed source, the garment; the model is the tool's to invent. A try-on has two fixed sources, the garment and the person, plus the person's photograph as the scene. The garment checks are the same; the person, removal and join checks are added, and they come first.

What this page does not claim

That any tool passes this checklist at a given rate, or that try-on images which pass it sell more. It is a list of what can be wrong when the wearer is fixed and how to see it; the rate is something you measure on your own garments and your own wearers.

To dress a garment on a model you photographed and kept, see virtual clothing try-on; for the fit rows in depth, how to preserve garment fit in virtual try-on; for the garment rows, how to prevent AI from changing clothing details.

Companion guides