How to Make Apparel Video Ads That Show Fit and Fabric

Fit and drape cannot be faked well, and this is the limitation to plan the whole creative system around. Generated clothing sits on the body like a texture, not like fabric with weight, grain and stretch. Shoulder seams float, hems do not break, and knits do not cling and release as a person moves. Do not generate a model wearing your SKU and ask media to do the rest. Use the avatar off the garment and cut to real on-body footage for every fit claim.

Drape is a physics problem: gravity, fabric weight, grain and the body's pressure points. Avatar engines are built to move a face and a torso, not to simulate how a 200gsm twill folds when someone sits down. The model guesses a plausible garment shape, then locks it, which is why you get repeating wrinkles, glued hems and cloth that ignores the knee. Close-ups make it worse. If a shot only works because the cloth moves honestly — a skirt in a turn, a blazer as the arm lifts — that shot has to be filmed. There is no prompt that repairs this yet.

Exact-SKU generation fails for the same reason plus logos, stitch lines and colourways. A buyer who opens the PDP and sees a different pocket placement will bounce, and that bounce does not show up as a creative metric. Structured outerwear and stiff denim hide sins better than silk, satin, rib knits and anything bias-cut. Stretch activewear looks painted-on; sheer layers look muddy. If your hero product is a fluid dress, budget for real movement footage or the ad will under-perform no matter how good the hook is. Accessories are kinder: bags, jewellery and sunglasses still need macros, but they do not depend on cloth hanging off a moving body.

You need on-body assets, not necessarily a studio. A well-lit phone shoot with one person who actually fits the sample size beats a generated model every time. Capture a standing front, a three-quarter turn, a sit-down, a fabric pinch, and one walk-away. Shoot vertical, handheld, close enough to see cloth move. That library is what makes every later script convertible.

Film eight to ten clips at eight to twelve seconds each even though a 30-second ad will only use six to eight. Extra length lets you cut on the sit, the arm-raise, the pinch — the moments that answer the sizing objection. Say the sample size and the model's usual size out loud on at least one take: I am 5'6, usually an 8, this 8 is true to size. Keep lighting consistent so the intercut with the avatar feels like one room. Hands, zippers and buttons are the same failure cluster as every other fine-motor demo; film the dressing moment even if it is only two seconds of a real hand on a real zip.

Save clips tagged as hook, fit, motion, fabric and packaging, and reuse them under new hooks and new avatars. Fatigue still applies: the same three-quarter turn in every retargeting ad at frequency 2.5–3.5 will start to feel like one creative. Keep at least three interchangeable versions of the walk and the sit-down. One afternoon of filming can feed a month of 6-creative test batches if you rotate the visual, not only the first line. Studio work still wins for the PDP. For Meta and TikTok prospecting, handheld, slightly imperfect on-body video is the asset that makes the avatar script convertible.

Sizing is apparel's biggest pre-purchase question and the biggest stated reason for returns — answer it in the script, not only on the size chart. Keep it specific and boring: construction, inseam, rise, stretch, whether it is lined, whether it runs small, that returns are free. Do not promise a perfect fit or a life-changing silhouette. The ad's job is to make the chart feel trustworthy, not to replace it.

A pretty generated body in a perfect drape will increase orders and returns together, which is a loss once you count shipping. Hybrid ads that show sitting, raising arms, and a size callout tend to attract fewer surprise-fit buyers. Medicalised slimming language, before-and-after body shots, and loses-two-sizes claims get fashion ads rejected even when you think you are selling clothes, not a waist trainer. Stay with construction. If a line would look wrong on a hangtag, it is wrong in the voiceover. The advertiser owns compliance; generation does not certify a given claim.

Try-on is an editorial format, not a sizing engine. Cutting a real person changing, overlaying size callouts, or showing two sizes on the same real body is legitimate when the footage is real. Generating a shopper's body into your garment and presenting that as proof of fit is how you create returns and policy risk. An avatar cannot be a size 8 or a size 18 in any way that maps to your grade. Formats that convert — GRWM, outfit from three angles, what I ordered vs what arrived, a 10-second wear-to-work loop — all need real cloth on a real body for the garment shots. The arrived state must be the real product, same colourway, no beautifying filter that changes the drape.

Cast for the customer three years into wearing the clothes, not for a runway. Match age band, energy and setting — bedroom mirror, walk-in, car on the way out — then test two faces rather than falling in love with a preview. Keep the avatar in a mirror selfie, a seated bed-shot or a cropped standing talk. Put full-body proof in your uploaded footage. Then cut as if the garment were the star.

A working 30-second cut is roughly 12 seconds of talking head and 18 seconds of on-body product, not the other way around — about 40% avatar and 60% real product for garments, closer to 60/40 for accessories. Fashion is a demo-led visual category even when the script is a review. If hold rate sits below about 12% while hook rate is above 30%, the cut is talking too much and showing too little cloth. If hook rate is weak, the opening is the problem, not the mix. Never open on a talking head with no garment in frame for this vertical — show the product or the outfit in the first second, then let the avatar comment.

Exact matching to the PDP model is impossible and not the job: the PDP model is wearing the SKU, the avatar is not. What you want is compatible age, styling and setting so the cut does not feel like two campaigns glued together. Avatars are built for waist-up. Full-body walking, turning on a spot and runway stride slide into uncanny output people screenshot. If a concept only works as a full-body walk, film that concept. Do not force the avatar to do it. Size-inclusive messaging needs real samples on real bodies in those sizes; token casting without matching garment footage is visible immediately.

One hero SKU, one body script that names size and fabric, 3 hooks × 2 avatars, all 9:16 at 1080p, one consolidated ad set. Hooks worth trying first: outfit already on frame, size-objection, ordered-versus-arrived. Export 9:16 first; add 1:1 only after something wins. Then judge the winner on return rate after the window closes, not only on platform ROAS.

Use the same diagnostic stack as every vertical, in order: hook rate target 30%+ (weak below ~20%), hold rate healthy at 12–20%, cold CTR target 1%+, then CPA against contribution margin after expected returns. Fashion often shows a decent hook on outfit-in-first-frame creatives and a soft hold if the garment never moves. Read tests at 48–72 hours or about 1,000 impressions per variant. Do not scale a winner until you have return-rate data, not just platform ROAS. A cheap CPA on a 40% return SKU is not a winner. If returns climb as CPA falls, the creative is overselling the garment. Kill that angle even if hook rate looks elite.

On cold audiences, fatigue typically shows at frequency 2.5–3.5, and fashion is on the early side because outfits are visually memorable. Refresh the opening outfit or the first line before you rebuild the body script. Swap avatar second, body copy third, offer last. Weekly 6-creative batches beat monthly campaign films. If both avatars lose and hooks also lose, the offer or the product shots are the problem, not the tool. When the product only makes sense in motion you cannot film — couture drape, tailored suiting sold on canvas structure — a human creator still wins. Use generation as the script-and-hook factory around footage you already have, not as a replacement for showing the clothes.