What changed in 2026
Ghost mannequin photography used to mean two things at once: a shoot, and then a retouch. You photographed the garment on a mannequin or a form, shot the neck and cuff interiors separately, and a retoucher combined the plates so the garment kept its worn volume with nothing visible inside it.
The retouching half of that can now be generated. You upload a photo of the garment worn, on a hanger, or laid flat, and the model returns an invisible-mannequin plate with the interior opened up. It takes about twenty seconds instead of a day.
That does not make manual retouching obsolete. It changes which jobs each one is right for, and the honest answer is that most catalogs need both.
How AI ghost mannequin actually works
The interesting problem is not removing the person. Removing a body is comparatively easy. The hard part is that a garment photographed on a body is full of information about how it hangs, and a garment photographed flat has almost none.
Three things have to happen for the output to read as a real ghost shoot rather than a cut-out:
- The garment has to stay three-dimensional. Shoulders need to hold a cap, sleeves need to stay tubular, and the chest needs volume the fabric is draped over. A flat silhouette with the middle erased is the most common failure and it is instantly recognisable.
- The interior has to open up. A ghost mannequin reads as one because you can see into the neck, and sometimes the cuffs and hem. That interior is not in the source photo when a body is filling it, so it has to be reconstructed plausibly.
- Construction has to survive. A seam is a physical join, not a drawn line. Yokes, plackets, dart lines and topstitching all have to land on the same construction lines they occupied in the original, while following a new drape.
Back views are the strictest test of all three, because there is no placket or graphic to anchor to. A centre-back seam, a yoke and a label are the only landmarks, and the model has to resist inventing a front.
Where AI wins outright
Volume and turnaround. Per-image cost falls to a fraction of a manual retouch, and the result arrives in seconds. For a 500-SKU seasonal drop where the shots are conventional garments in good light, this is the entire argument.
Garments you never shot properly. If all you have is an on-model image from a previous season, or a supplier hanger shot, there is no plate to retouch. There is nothing for a retoucher to combine. Generation works from what exists.
Cheap iteration. Changing the background from white to grey, or trying it with and without a contact shadow, costs another generation rather than another brief and another round trip.
Sampling before you commit. Running a handful of SKUs through in an afternoon tells you whether a format works for your catalog before you book anything.
Where manual retouching still wins
This is the part most tool comparisons skip, so here it is plainly.
Hero and campaign images. If an image is going on a homepage, a paid campaign or a wholesale line sheet, it is being looked at properly. Generated output is very good and occasionally not exact. A retoucher is exact.
Complex or unusual construction. Pleating, heavy structured tailoring, sheer layers over opaque ones, unusual closures, elaborate draping. Anything where the correct answer is not inferable from the photograph is where a human who can see the actual garment wins.
Colour-critical work. A precise fabric colour against a brand standard, or a fabric with a difficult finish, is a measurement problem rather than a plausibility problem.
Interiors that must be literally correct. Generated interiors are plausible reconstructions. If the exact printed lining or the exact brand tag has to appear, supply a reference or retouch it.
The workflow most catalogs actually want
Split by what the image has to do, not by what is newest.
- Generate the long tail. Standard garments, standard shots, high volume. This is the bulk of any catalog and where the cost and speed difference compounds.
- Review at a glance. Check the neck interior, the shoulder line, and the seams against the source. Most of what needs fixing is visible in seconds.
- Send the exceptions to a retoucher. Hero shots, difficult construction, colour-critical pieces, anything that failed review.
The proportion depends on your catalog. A basics brand might generate almost everything. A tailoring or occasionwear brand will send far more to manual. Both are correct answers to different catalogs.
Cost and speed, side by side
| AI ghost mannequin | Manual retouching | |
|---|---|---|
| Turnaround | About 20 seconds per image | Typically 24 hours |
| Per-image cost | One generation credit | From $0.49 to $2.50 |
| Input needed | Worn, hanger, flat lay or packshot | Shot plates, ideally with interiors |
| Exactness | Plausible reconstruction | Pixel-exact to the garment |
| Best for | Volume, long tail, iteration | Hero shots, complex construction |
| Revisions | Regenerate, seconds | Brief and round trip |
What to check before you publish
A fast review pass catches nearly everything worth catching:
- Shoulder line. Does it hold a shoulder shape, or has it collapsed flat?
- Neck interior. Does the opening read as depth, or as a hole cut in fabric?
- Seam continuity. Do yokes, plackets and side seams land where the original has them?
- Sleeve volume. Are sleeves tubular, or flattened into paddles?
- Fabric colour. Compare against the source, not against memory.
- Hardware. Button count, zip pulls and trims all present and in the right places.
Frequently asked questions
Is an AI ghost mannequin image good enough for Amazon or a marketplace?
For standard garments, generally yes. Marketplaces care about a clean background, the garment filling the frame, and no visible model or mannequin, and a generated plate meets those the same way a retouched one does. Check the specific listing spec for your category and review the image before it goes live.
Can AI ghost mannequin produce back views?
Yes, and it is worth requesting them explicitly. A back view has fewer landmarks than a front, so specifying a back shot keeps the centre-back seam and yoke in place and stops the model inventing a front placket.
What about linings and inner labels?
Generated interiors are plausible rather than literal. If the exact lining or brand tag matters, supply a reference image of it, or send that SKU to manual retouching.
Does AI ghost mannequin work from a flat lay?
Yes. A flat lay is a harder input than a worn shot because gravity and drape have to be inferred rather than observed, so expect a slightly higher review rate from flat-lay sources on structured garments.
Do I have to disclose that a product image was AI generated?
It depends on the channel and on where you sell, and guidance is still moving. Treat it as a policy question for each platform you list on rather than assuming one answer covers all of them, and take proper advice if you are unsure.