PDP UGC is the SKU on the page

PDP UGC is the SKU on the page. The pack still has to match the product-detail photo the shopper can already see. Image-to-video from that still. A different bottle, a different label colour, a “hero” angle that is not on the PDP is a different listing wearing this URL. The production constraint is the pack shot: readable type, the same SKU, the same dieline. The still contract is the rule that the clip must keep being that frame, not a prettier cousin.

This is not a lifestyle-brand film and it is not a marketplace-policy essay. It is the file that sits next to Add to cart: a short motion plate of the thing in the hero image. Hands are allowed. A kitchen is allowed. A new pack is not.

UGC ads that actually convert 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 on a PDP is someone holding the SKU the page sells, not a mood bottle. Human UGC still rates 81% on authenticity against 63% for AI UGC (Kapwing 2026; archive.com / creator.co). On a PDP the gap is the label. Shoppers pause on the word. They do not pause on melted type.

The pack still must match the PDP photo

Pull the hero still from the live PDP, or from the same DAM asset the PDP uses. Do not re-shoot a “better” pack in generate and hope the cart still feels like the same product. If the live page is the 12oz matte bottle and the clip is a 16oz gloss, you are advertising a sibling SKU.

Approve, at 100% zoom, before any motion:

Brand lockup, flavour or variant line, net contents, the claim the page already makes.
Cap colour, pump, spout, window — the silhouette a returning buyer uses.
Lot-adjacent marks you are allowed to show. Do not invent a batch.
Background. If the PDP is white, a marble slab in the clip is a different set. Lifestyle can be a second module. The pack module has to be the pack.

If the still would fail a pack-shot audit, motion will not save it. Retry the still. Image models are cheaper than video rows. Grok Imagine Image Quality is 4 credits when you need a still that will hold; the 2-credit Grok Imagine base is the loose 4K sketch. Neither is a reason to invent a dieline the PDP does not sell.

Do not:

Prompt a “premium redesign” of a live SKU.
Swap the variant because the mango label is prettier than the SKU in the URL.
Add a badge the page does not show (awards, “#1,” a percentage you cannot document).
Mirror the pack so type reads wrong.
Crop the legal line off and hope the clip feels cleaner.

Marketplace and brand.com PDPs both count. If the ad lands on Amazon, the clip has to survive that gallery. If it lands on your site, it has to survive the hero. Two URLs, two stills, if the photos differ.

Image-to-video from that still, not a new pack

The job is I2V. Frame one is the approved PDP still. The prompt names what moves: a hand, a pour, a lid, a slow push, a turntable. The prompt does not re-describe the product. Re-describing the box is how the model tries to satisfy a sentence that is already true and walks the type off the dieline.

Rows that actually take a still:

Wan 2.6 Image to Video — 20 credits, 1080p, 5 / 10 / 15 seconds, native audio, image required. Hero motion when you want room sound or a short line.
Wan 2.6 Image to Video Flash — 20 credits, 1080p, same lengths, audio false. Mute draft, or mute ship if VO will be burned later.
Kling O3 Standard Image to Video — 45 credits, native audio, 3–15 seconds. Reasoning on the motion, not a new bottle.
Flux 3 First Last Frame to Video — 34 credits, native audio, 5–20 seconds, when you have a start still and an end still of the same SKU.

Do not start the PDP module on a text-to-video row. Veo 3.1 at 64 credits will invent a handsome bottle in 4 / 6 / 8 seconds. Handsome is a different ASIN. Grok Imagine Video at 24 credits will give you 6–30 seconds at 720p from a sentence. A sentence is not the PDP photo.

Native audio is a flag, not a licence to talk over a melted label. If the row can speak, write whether anyone speaks. SKU orbits usually want foley and no voice. Say “no speech.” A surprise monologue on a pack still is a wasted take.

Duration: pick the length the PDP module can hold. A 15-second orbit on a pdp-hero that autoplays muted is often too long. Five or six seconds of honest motion plus a hold on the still is a module. Cutting a 15-second take down is fine; paying for 15 because the slider felt adjacent is not.

Hands, lifestyle, and the second module

UGC on a PDP is allowed to leave the white sweep. The second module is hands, pour, wear, open — still the same SKU. If the hands put a different variant on the table, you have two ads.

Hands: adult, real, on the pack you sell. Pick a human when the hands matter is the pump, the tear, the scoop. Do not generate a cream smear or a pour that implies a viscosity the formula does not have. Do not generate a child handling the SKU.

Lifestyle sets are not the pack shot. A kitchen can be a kitchen someone has. A generated marble island with a generated bottle is two lies. Film the kitchen or skip the kitchen. Keep the pack as the still you already approved, even if you cut away.

Claims stay on the PDP. The clip does not get to add a disease line, a “clinically proven” badge, or a percentage the page does not show. A creator VO is still an ad claim. The FTC’s Consumer Reviews and Testimonials Rule still applies to a handheld take. A generated “this changed my skin” is a fabricated testimonial ad.

Captions: burn the variant name as the PDP names it. Keep the block off the label. Covering the proof with type is a unique way to waste a pack still. For mute autoplay, write the one line the shopper needs if they never unmute — see captions for silent autoplay. The AI caption generator is for that line, not for a new claim.

Article 50 wants a first-exposure label on photoreal people for EU audiences since 2 August 2026. A generated hand-model is a person. A labelled fake reviewer is still a fake reviewer. New York’s synthetic-performer law (N.Y. GBL § 396-b, in force 9 June 2026) wants conspicuous disclosure of generated human performers in ads that can reach New York consumers. $1,000 first, $5,000 subsequent. Disclosure does not make the SKU match.

One SKU, one URL, one still

A size set is not one PDP. 12oz and 16oz are two stills if they are two URLs. A flavour set is not one PDP. Bundle pages need the bundle still, not a stack you invented in generate.

Colourways: the clip is the colourway on the page. A carousel of every colour is a catalog ad. The PDP module is this colourway, this size, this scent word.

If the live photo is out of date — old label, old cap — fix the PDP first. Animating a retired dieline is how returns start. The still contract includes “this is the pack we will ship this week.”

The UGC video generator can wrap the I2V plate with a creator VO you recorded and burned-in variant names. Do not prompt it for “a cinematic bottle on a cliff.” Prompt it to move the still the PDP already shows.

66% of creators offer UGC services and the creator population grew 93% year on year (Kapwing 2026; creator.co). Most of them should be handed the live pack, not a moodboard. 87% of brands use UGC and 16% have a strategy. The strategy on a PDP is: same still as the hero, I2V from that still, hands if they help, no new bottle.

If the clip still works after you swap the URL, you did not lock the SKU. Put the PDP photo in the upload slot.

Move the live pack still in Versely.

FAQ

Can I use a lifestyle generate if the label is readable?

Not as the pack module. Readable type on a bottle the PDP does not sell is still a different SKU. Lifestyle is a second module, and the bottle in it still has to be this bottle.

Which model should I start on?

Start on an image-to-video row with the live still. Flash at 20 credits is the mute draft. Wan 2.6 I2V at 20 credits is the native-audio hero on the same lengths. Do not start on Veo text-to-video.

What if the PDP photo is a bad still?

Fix the still. Retake or recrop until the label holds at 100%. Then animate. A 15-second clip of a soft label is a long, confident error.