No filming:An AI avatar or clone yourself
    Guides

    Fashion UGC is fabric under real light

    Fashion UGC is fabric under real light: drape, size, and colour. Still the garment, then I2V. Fake models plus fake cloth become returns.

    Versely Team••7 min read

    Fashion UGC is fabric under real light. Drape, size, and colour have to belong to the unit you will ship. A generated model in generated cloth is a lookbook of a garment that does not exist, and that file becomes a return when the hem, the stretch, and the dye do not match the PDP. Photograph the garment. Then, if you need motion, run image-to-video from that still. Do not text-to-video a dress and hope the SKU catches up.

    The production constraint is cloth. Knit versus woven, bias versus straight, a waistband that cuts or does not, a black that is blue in daylight. Viewers who buy clothes already know when a fabric has no weight. Comments will say “that’s not how that twill moves” before they say they want a size.

    This is not beauty shade-matching and it is not a catalog CGI pipeline. It is a peer wearing, holding, or hanging a specific SKU in a room that has a window.

    UGC ads still show about 4× click-through and about 50% lower CPC, and 86% of brands say UGC lifts performance (Kapwing 2026; Billo). 92% of consumers trust a peer over a brand message in the same stack. A peer in fashion is someone whose body and whose light you can compare to yours. Human UGC still rates 81% on authenticity against 63% for AI UGC (Kapwing 2026; archive.com / creator.co). The gap is drape.

    Drape under a window, not a prompt

    Drape is gravity plus fibre. Models that generate cloth tend to invent a fabric with no mass, or a mass that never wrinkles at the elbow. Either tell is enough. Real light on real cloth shows the third: colour shift. A navy that goes purple at the window is information. A grade that crushes it back to hex #000080 is a lie about the dye lot.

    Film:

    • The garment on a hanger, then on a body, then in a hand at the seam. Three weights. Hanger shows the cut. Body shows the size. Hand shows the thickness.
    • One window or one overhead, not both fighting. Mixed colour temperature is how black becomes two blacks.
    • A walk that is a walk. Four steps, turn, sit if the SKU is meant to sit. Slow-motion fabric ads are lookbooks. UGC can be a lookbook if you stop calling it a peer review.
    • The label, care tag, and SKU code if they are part of the proof. Size on the tag should match the size the caption claims.

    Do not generate a breeze. Do not generate a catwalk. Do not generate a body that is not the talent you hired, wearing a garment you did not send them.

    Pick a human when the hands matter applies to seams, buttons, zips, and the way a cuff actually sits. Hands on a zip are a size story. Generated zips that never catch are a different product.

    Size and colour are return reasons

    Fashion returns cluster on “not as pictured.” UGC that is not as shipped is an ad for a return. Write the size the talent is wearing, the size they usually wear, and one measurement they took (rise, chest, length). Do not say “fits true to size” unless you have a size run in the same light and you are willing to show more than one body.

    Colour: name the colourway as the PDP names it. Auto-captions will say “blue.” The AI caption generator is for burning “Bone” or “Ink 02,” not for synonyms. Photograph a colour card or the pack card in the first seconds if the dye is the reason to buy.

    Do not:

    • Stretch a generated body to imply the garment slims.
    • Crop a hem that is short into a hem that is not.
    • Run a filter that shifts the colourway toward last season’s hero.
    • Mix two colourways in one file without labelling the cut as a lineup.

    A size run is several ads or a single clearly labelled sequence: same room, same light, three bodies, three tags. Generating the two bodies you did not cast is how you invent a fit. Cast the two bodies. Send the garment. Film the tags.

    Minors: do not generate child models. Children’s apparel, if you sell it, is a real child with parental consent and platform child-safety rules, or it is a hanger still. This page does not use generated children as a size reference.

    Still the garment, then image-to-video

    The still is the SKU contract, same as a beauty pan. Lay the garment, hang it, or photograph it on the talent. Then animate that file.

    Image-to-video from a true still keeps the stitch count, the print registration, the button colour. Text-to-video invents a print that almost matches. “Almost” is a counterfeit read, even when you own the brand.

    Seedance 2.0 is a 4–15s native-audio row at 1080p (from 44 credits) that will happily invent a model in a dress. Use it on a reference still of the actual garment if the object has to survive, or use it for a clearly labelled graphic world that is not the PDP. Veo 3.1 will give you a cinematic 4–8s cloth simulation from a prompt; cinematic cloth is not your dye lot.

    When I2V is the right job:

    • A hanger still that needs a slow rotate.
    • A on-body still that needs a small weight shift, not a new body.
    • A detail still (stitch, hardware, print) that needs a push-in.

    When it is the wrong job:

    • You do not have the garment yet.
    • You want a second colourway you have not sampled.
    • You want a model whose measurements you do not have, because the sample only fits one person.

    Unboxing of apparel is tissue, a card, a fold, a smell-of-the-box moment. Fine as a shipping ad. It does not replace drape. An unbox that never puts the garment in light is not a fit ad.

    Fake models plus fake cloth

    Stacking a generated body and generated cloth is a campaign of a fictional SKU. AI UGC can still do talking openings — “this is the jacket I actually wear on the train” — if the jacket in the cutaway is the jacket, and if you do not imply the generated face is the wearer unless you disclose and you accept the authenticity drop (81% human versus 63% AI in the Kapwing 2026 stack). The safer split is a real wearer and generated hooks around their footage.

    New York’s synthetic-performer law (N.Y. GBL § 396-b, in force 9 June 2026) requires a conspicuous disclosure on ads that use a generated human who is not an identifiable person, when you have actual knowledge and the ad can reach New York consumers. First penalty $1,000, subsequent $5,000. That is a label. It does not fix the return.

    Article 50 wants a first-exposure label on photoreal people and places for EU audiences. Use the Article 50 disclosure template. A labelled fake coat is still a fake coat.

    66% of creators offer UGC services and the creator population grew 93% year on year (Kapwing 2026; creator.co). Send samples. Get them back. 87% of brands use UGC and 16% have a strategy. The strategy here is: one colourway, one size labelled, real light, still then I2V, no invented drape.

    If the cut still works after you swap the fabric in post, you never had the garment. Put the real hem in the window.

    Still the garment in Versely.

    FAQ

    Can I generate a model if the garment still is real?

    You can composite a disclosed presenter next to a hanger still. Do not drape generated cloth over a generated body and call it that SKU. If a body is wearing it, that body received the unit.

    Why not film in a studio with a cyclorama?

    You can. That file is a lookbook. UGC in this category is a window and a walk. Studio light hides drape that a hallway bulb will reveal on the day the parcel arrives.

    Do I need a full outfit?

    No. One SKU. Styling pieces in the background should be out of focus or rights-cleared. A full look of five unsold items is a campaign, and it confuses the PDP you are pointing at.