Catalog product video examples. Same photo, two prompts.
Seven invented products, one reference image each: two takes from a careless prompt, two from a locked one. Every prompt is printed as it ran.
Tell us what you sell. We'll come back with which SKUs should get video first.
Free. No card. We reply within one working day.
In our tests, AI catalog video went wrong in four ways: the shape changed, the color shifted, text appeared on the label or the size drifted. Each starts when the model invents what one photo did not show. Across 15 invented products, a prompt locking parts, color, size and "no text" held in 26 of 30 takes.
- Drift shows up when the camera, the light or the set asks for something the photo never showed.
- Write down the part count, the material, the color as a hex and the size in centimeters.
- Tell the model the image wins where the text disagrees.
- Ban generated text. Add names and prices afterward as a layer.
- Watch every take: 4 of 30 locked takes still missed.
Last updated · QuickAds Editorial Team
Every product is invented. Each pair used the same reference image, engine and length, and only the prompt text changed.
What do catalog product video examples look like? Seven products, two prompts each.
Each card holds the careless take, the locked take, the reference image both came from and both prompts as they ran. Every prompt ran twice, so each tally counts two takes.
Golden hour turned a red tee orange
Golden hour recolored the fabric: brick in the shade, orange in the sun, 19 ΔE from the red we sent (ΔE scores how different two colors look). The locked prompt gave the hex and banned warm light, and the locked take shown sits at 4 ΔE.
Careless prompt
Locked prompt
Reference image sentRed performance t-shirt on a model, dynamic movement, golden hour lighting
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the garment; sample the fabric color from it and hold it: deep sundried-tomato red, midtone #892128. Lighting must not recolor the fabric. Neutral daylight, no warm gels, no golden hour. Model turns 20 degrees left over 4 seconds; fit and sleeve length unchanged. Negative: orange, burgundy, magenta, color cast on fabric.
A warm kitchen made a red kettle maroon
Steam and a warm kitchen pulled the signal red to maroon, 24 ΔE off for all eight seconds. The locked prompt named the hex and neutral daylight. One locked take held within 4 ΔE, and the other came back a stop darker.
Careless prompt
Locked prompt
Reference image sentRetro red kettle on a stove, warm kitchen morning, steam, cinematic
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the product; sample the enamel color from it and hold it: signal red, midtone #B3191F. Lighting must not recolor the enamel. Neutral daylight, no warm gels, no golden hour. Camera: slow push-in over 6 seconds on the kettle standing on a gas hob, no steam. Negative: orange, crimson, pink, color cast on the enamel, extra kettles, resizing.
A blank label came back with a real brand's name
We sent a shampoo bottle with a blank label. Both careless takes printed a brand name on it, and the name belongs to a real company, so we blurred it. The locked prompt banned all text, and the label stayed blank.
Careless prompt
Locked prompt
Reference image sentShampoo bottle with brand label rotating on a white background, product commercial style
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the product. Do not render or redraw any text or logo; keep the label area as a flat neutral patch. Camera: 90 degree turntable over 6 seconds, white sweep, soft top light. Negative: text, typography, letters, logo, watermark.
The tote grew about a third between seconds four and six
The careless take shown kept the right size for four seconds. By the sixth it was about a third wider and taller. The locked prompt gave the size, 38 by 30 cm at hip height, and the tote kept it through the walk.
Careless prompt
Locked prompt
Reference image sentWoman walking with a stylish tote bag, fashion film, slow motion
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the bag. Medium tote, 38 cm wide, 30 cm tall, sits at hip height on the shoulder; not oversized. Natural weight: the bag swings slightly with each step and the strap stays the same length. Model walks toward camera, three steps, full bag in frame, nothing cropped. Negative: oversized, floating, strap stretching, bag resizing.
A fine chain got three times heavier in the close-up
In one careless take, the 1.2 mm cable chain turned into a paperclip chain three times the weight when the camera cut in close. The locked prompt gave every size as a number and said none of them changes at any frame.
Careless prompt
Locked prompt
Reference image sentElegant necklace on a model, close up, jewelry ad, sparkle
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the product; reproduce it exactly. Fine 1.2 mm yellow gold cable chain, 45 cm, with one 8 mm round polished disc pendant, worn on bare skin at the collarbone. Chain thickness, pendant size and length do not change at any frame. Model breathes and turns 10 degrees; the pendant settles with natural weight. Neutral daylight. Negative: thick chain, medallion, oversized pendant, extra pendants, stones, clothing at the neckline, sparkle effects, resizing.
The rattan weave rebuilt itself as the camera moved
On a slow orbit the rattan weave never settled: strips wide apart at two seconds, dense at seven, new ribs in between, the dome flattening. The locked prompt gave a 40 cm dome of evenly spaced strips and held the rib count to the reference image.
Careless prompt
Locked prompt
Reference image sentA rattan pendant light in a cozy living room, camera slowly orbiting, warm lamp glow, cinematic, 4k
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the product; reproduce it exactly. Pendant shade woven from natural rattan, a 40 cm dome, an open weave of evenly spaced horizontal rattan strips lashed to vertical ribs, one plain black cord entering the top, no bulb visible. Rib count, weave spacing and dome silhouette match the reference image and do not change at any frame. Camera: slow 30 degree orbit right, eye level, no push-in. Soft even daylight, plain background, shade unlit. Negative: extra ribs, changing weave, closing weave, tassels, beads, fabric shade, second lamp, morphing.
One label band split into two floating patches
One careless take split the jar's wrapping label into two floating patches and gave the lid a knurled rim. The label stayed blank, so only the parts changed. The locked prompt asked for one label band at every frame.
Careless prompt
Locked prompt
Reference image sentLuxury skincare jar rotating on marble, beauty commercial, soft glow, premium branding
Reference image is the product; where this text and the image disagree, the image wins. Reference image is the product; reproduce it exactly. Squat amber glass cosmetic jar with a brushed gold screw lid and one continuous matte cream label band wrapping the body, with nothing printed on it. One label and one only, a single band that wraps the jar at every frame; the lid rim stays smooth and plain, with no knurling and no ridges. Part count and part shapes do not change at any frame. Camera: 90 degree turntable over 6 seconds, white marble, soft top light. Negative: second label, split label, floating label patch, knurled lid, ridged rim, text, lettering, logo, embossing.
The careless prompt did not break these eight
Eight of the 15 products came through the careless prompt intact in both takes, and four of those were material tests.
- The oak dining chair kept four legs and two slats through a 12-second orbit.
- The pebbled top-handle bag held its grain and antique brass through the pan.
- The 25 mm gold hoops stayed one hoop per ear.
- The trench kept ten buttons, both epaulets, the storm flap and the belt through a walk.
- The suede Chelsea boot kept its nap under a dark key light.
- The brushed steel bottle stayed brushed through a push-in, with no slide toward chrome.
- The cable-knit sweater kept both cable columns through golden hour and a turn.
- The 38 cm velvet cushion kept its size. The room changed instead: a three-seat sofa came back with two.
Why does AI catalog video drift? Five failure classes, four seen in our tests.
AI catalog video drifts when the model has to invent what the photo did not show. Across 15 invented products, the careless prompt broke 12 of 30 takes and the locked prompt held 26 of 30. Material never broke.
| Class | What changes | The line that locks it | In our tests |
|---|---|---|---|
| Shape and geometry | Parts appear, vanish or change shape: new ribs in a weave, a label in two pieces, a second hoop. | Name and count every part, then add "Part count and part shapes do not change at any frame." | Careless: 3 of 10 broke Locked: 9 of 10 held |
| Material | Suede goes smooth. Brushed steel slides toward mirror chrome. | Name the exact material, such as "natural rattan." | Careless: 0 of 8 broke Locked: 8 of 8 held |
| Color | The light recolors the product: golden hour, a warm kitchen. | Give a hex, ask for neutral daylight and add "Lighting must not recolor the fabric." | Careless: 4 of 4 broke Locked: 3 of 4 held |
| Type | Letters appear on a blank label. Ours spelled a real company's name. | Add "Do not render or redraw any text or logo," with text and logo as negatives. | Careless: 2 of 2 broke Locked: 2 of 2 held |
| Scale | The product grows mid-clip or gets heavier in a close-up. | Give the size in centimeters or millimeters, and add "not oversized." | Careless: 3 of 6 broke Locked: 4 of 6 held |
Counts are takes, two per prompt per product. The velvet cushion is filed under scale, the lock it was testing.
Does a locked prompt always hold? Not every take.
Four of the 30 locked takes still drifted, against 12 of 30 for the careless prompt. One miss is enough to put the wrong kettle in a feed, so a person watched every take before any clip was published.
Hoops · a second hoop
Kettle · a stop darker
Cushion · the sofa grows- The hoop earrings prompt asked for one 25 mm hoop per ear. A second, smaller hoop sat inside the first for 9 of the 12 seconds, on a product the careless prompt had held in both takes.
- The red kettle take used the same prompt and hex as the kettle take that held, and rendered a stop darker: 7.2 ΔE off.
- In both velvet cushion takes, the room changed. In the one shown, a three-seat sofa grew to five seats by the sixth second. In the other, a second cushion came into frame on a lock that said one cushion only.
What people ask about AI product video.
Why does an AI product video change the product halfway through?
A photo shows one side of the product. Once the camera moves, the model has to invent the sides it never saw, and anything the prompt left open can change. In our tests, the changes were new ribs in a rattan shade, a jar label split in two, a fine chain turning heavy in a close-up and a tote that grew about a third.
What is the difference between a careless prompt and a locked prompt?
A careless prompt describes a mood: golden hour, cinematic, sparkle, premium branding. A locked prompt describes the product: it says the reference image wins where the text disagrees and gives the part count, the material, the color as a hex and the size in centimeters. A locked prompt also bans generated text and lists what must not happen. Both prompts in each of our pairs used the same reference image, engine and length.
How do you keep a product consistent in AI video?
Write the product down before the camera moves: each part and how many there are, the exact material, the color as a hex and the size in centimeters. Tell the model the image wins where the text disagrees, ban generated text and keep to one camera move. Then watch every take against the photo. In our tests, 4 of 30 locked takes still missed.
Can AI video hold my exact product color?
Mostly, when the prompt gives the color as a hex and keeps the light neutral. In our tests, 3 of 4 locked color takes held. A red tee measured 19 ΔE (a CIELAB color distance) from its reference under a golden-hour prompt and 4 ΔE under the locked one. A red kettle held in one locked take and came back a stop darker, 7.2 ΔE out, in the other. We measured color on the published files.
How do you check a clip before it goes in the catalog feed?
Watch the clip to the last frame against the product photo for shape, color, label text and size. For our published test clips, a person watched every take that way, and we measured color in CIELAB against the reference image, offline. QuickAds software does not measure color on video and has no approval step: its export writes every finished clip, so watch the previews before you export.
Are the QuickAds catalog video examples from client work?
No. We generated every clip on purpose, on products we invented, and no client product appears anywhere in the set. Because every product is invented, we can report every result, including the four locked takes that missed. Each of the seven pairs used one reference image, one engine and one length, so its careless and locked takes differ only in the prompt, and both prompts are printed with the pair.
Can you provide some examples of Meta catalog ads?
The QuickAds examples show only the video half of a Meta catalog ad. A Meta catalog ad reads each product's clip from a video field in its feed row as a direct file link, and, with Dynamic Media on, it can play that clip in the ad. Each example is one invented product made from a single reference image, 8 or 12 seconds long and silent, and the seven pairs print both prompts beside their clips.
One clip per SKU, across the whole feed.
QuickAds is a performance creative company that runs the whole chain: creative intelligence, creative strategy, creative production, influencer marketing and campaign management. Catalog product video is the link that gives each product row in your feed its own short clip for Meta catalog ads, and AI video ads cover the rest of your paid social. Pricing, the feed setup and what you send us are on the catalog product video page.
Meta reads each product's clip from a video field in the catalog feed, and that field has to hold a direct link to the file. Source