A ghost mannequin image has one job: to show a garment holding its own shape, with the inside of the collar visible and nothing inside it. When it fails, the shopper does not say why; they scroll. And when a brand regenerates the image without naming the failure, it gets the same failure with different details, because most of the mistakes were made before the mannequin came off, in the photograph, and the rest are particular to how the effect was generated.
This page is a field guide to ghost mannequin mistakes. It publishes no conversion figure and no error rate; it names the mistakes, the sign that identifies each at full size, and the fix that belongs to it. Half of the mistakes are the photograph's and are fixed by reshooting; half are the generation's and are fixed by changing the source or the setting, never by retouching the result.
What are the most common ghost mannequin mistakes, and how are they fixed?
In the photograph: a creased or badly dressed garment, a form the wrong size, uneven light with shadows in the armholes and under the collar, the inside of the collar not visible, a hem clipped by the frame, and a busy or matching background that gives the removal no edge to follow; each is fixed by reshooting, because no removal method recovers what the photograph does not show. In the generation: an inner collar invented because the source did not show it, a hem or a sleeve that changed shape, a garment a shade off, buttons and hardware regularised, a shadow that disagrees with the light, and a trace of the form left at a shoulder; each is caught by comparing the result to the photograph at full size and fixed by changing the source or the setting and generating again — never by painting over the result.
Two kinds of mistake
A photograph mistake is anything the removal method, human or AI, is given and cannot change: the garment's dressing, the light, the frame, what is visible. A generation mistake is anything the method did to the garment's pixels or added to them. They are separated by asking one question of a bad image — did the photograph show it? — and the answer decides the fix.
Mistakes in the photograph
- Creases and poor dressing. A garment steamed and pinned so it hangs as designed is the whole effect; a crease becomes a permanent feature. Fix: steam, pin, dress again.
- A form the wrong size. A garment stretched on a large form or slumped on a small one holds the wrong shape. Fix: a form in the garment's size.
- Uneven light, dark armholes. A shadow inside the armhole or under the collar becomes a hole the tool has to fill. Fix: soft, even light from the front and both sides.
- The inside of the collar not visible. The inner collar is the effect; if the source hides it, it is inferred. Fix: photograph the collar open and, for a traditional composite, the inner shot.
- A hem or sleeve clipped by the frame. What is not photographed is invented. Fix: the whole garment in the frame with margin.
- A background that matches the garment. A white shirt on white or a black jacket on a black form leaves no edge. Fix: a contrasting, plain ground.
- Hardware closed, pockets hidden. Buttons under a lapel, a zip hidden by a flap: a tool completes what it cannot see. Fix: show the hardware as the customer will see it.
Mistakes in the generation
- An invented inner collar. The lining, the label or the depth of the neck does not match the garment. Sign: compare the inner collar to the garment turned inside out. Fix: reshoot with the collar open; use the inner shot where the workflow takes one.
- A changed silhouette. A hem straightened, a sleeve narrowed, a shoulder squared. Sign: the outline against the photograph at full size. Fix: a source that shows the shape clearly; one change per image.
- A colour shift. The garment a shade warmer or cooler than the photograph. Sign: the grey card. Fix: a neutral ground and a card in the source; regenerate.
- Regularised hardware. Buttons evenly re-spaced, a zip drawn as a line, a pocket added for symmetry. Sign: counted against the photograph. Fix: hardware visible and sharp in the source.
- Texture smoothed or sharpened. A knit turned into a flat surface. Sign: the weave at full size. Fix: full-resolution upload — up to 10 MB in Photta — and a sharp source.
- A shadow that disagrees. A contact shadow from the wrong side or two shadows. Sign: the light on the garment versus the shadow. Fix: the studio setting that matches the source's light.
- A trace of the form. A grey edge at a shoulder or an armhole. Sign: the edge at full size. Fix: a contrasting form and ground; regenerate.
The fix that is never right
Painting over a generated mistake — a brushed-in collar, a cloned button, a straightened hem — produces an image that is neither the photograph nor the generation, and the customer receives the garment the photograph shows. Every fix on this page is a change to the source or the setting followed by a new generation, and a comparison afterwards.
Photograph, compare, trace
The loop the checklist below serves, named so the other ghost pages can point at it.
- Photograph — the garment dressed, lit, fully framed, collar open, on a contrasting form and ground, with a grey card.
- Compare — every result beside its photograph at full size, part by part.
- Trace — each failure to the photograph or to the generation, and fixed there: reshoot, or change the setting and generate again.
Ghost mannequin mistake checklist
| Look for | Kind | Sign at full size | Fix |
|---|---|---|---|
| Creases, slumping, stretch | Photograph | Present in the source too | Steam, pin, right-size form, reshoot |
| Dark armholes, collar shadow | Photograph | A filled or blurred hollow | Even light from front and sides |
| Inner collar unseen in source | Photograph | Rendered collar unlike the garment | Collar open; inner shot |
| Hem or sleeve clipped | Photograph | Invented edge | Whole garment in frame |
| No edge to cut | Photograph | Halo or lost edge | Contrasting ground and form |
| Invented inner collar | Generation | Lining or label the garment lacks | Reshoot collar open, regenerate |
| Changed silhouette | Generation | Outline differs from the source | Clear source, one change per image |
| Colour shift | Generation | Card differs | Neutral ground, card, regenerate |
| Regularised hardware | Generation | Counts differ | Hardware sharp in source, regenerate |
| Texture altered | Generation | Weave differs at full size | Full-resolution sharp source |
| Shadow disagreement | Generation | Shadow against the light | Match the studio light to the source |
| Trace of the form | Generation | Grey edge at shoulder or armhole | Contrasting form, regenerate |
Every row ends in a reshoot or a regeneration; none ends in a retouch.
One range, one review
A shirting brand runs a range of forty shirts through the ghost mannequin generator from one photograph each and reviews every result beside its source. Nine are rejected. Four are photograph mistakes — two creased fronts, an armhole in shadow, a hem clipped — and are reshot. Five are generation mistakes: two inner collars showing a contrast lining the shirts do not have, one placket with the buttons re-spaced, one shade shift on a pale blue, one grey trace at a shoulder. The collars are reshot open, the placket source is sharpened, the blue is regenerated with the card in frame, the shoulder is re-run on a darker form; none is painted. The forty images that go live show the shirts in the box.
Frequently asked questions
Why does my ghost mannequin image look wrong when the garment is fine?
Usually because the photograph did not show something the method had to supply: the inside of the collar, a shadowed armhole, a hem outside the frame. Compare the result to the photograph; if the photograph lacked it, reshoot; if the photograph had it and the result changed it, change the setting and regenerate.
Can I fix a bad ghost mannequin image in an editor?
Not honestly. Painting in a collar or a button produces a garment that is neither the photograph nor the generation. Fix the source or the setting and generate again, then compare.
What is the most common AI ghost mannequin mistake?
An invented inner collar — a lining, a label or a neck depth the garment does not have — because the source did not show the inside. Photograph the collar open and check the rendered interior against the garment turned inside out.
How should I photograph a garment for ghost mannequin?
Steamed and pinned on a form in its size, on a contrasting plain ground, under soft even light from the front and both sides, the whole garment in the frame, the collar open, hardware visible, a grey card at the edge, at the camera's full resolution.
How do I check a batch quickly?
Compare each image to its photograph at full size, name the kind of failure — photograph or generation — and group the rejections by row. A batch that fails mostly in one row has one fix, and the next batch fails less.
What this page does not claim
Any error rate for any method, any effect of a fixed image on conversion, or any time a fix takes; this page measures none of those. What it claims is that ghost mannequin mistakes are made in the photograph or in the generation, that each is identified by comparison and fixed at its source, and that a retouch fixes nothing.
To generate ghost mannequin images from one photograph, see the ghost mannequin generator; for the full production sequence, how to create realistic ghost mannequin images step by step; for shooting on a phone, 3D ghost mannequin photography with a smartphone.