A standing multi-SKU lineup in AI video is a row of distinct finished SKUs. The job is to stop the model from merging them, swapping sizes, or inventing a third product. An overhead flat lay is a different page. Use the flat lay workflow when the camera looks down on a kit already spread on a surface. This page is for upright, finished products a shopper should be able to point at and name.
The lineup still is the authority. Image-to-video draws new frames from that still, and products that stand near each other are easy to blend. A careful prompt can limit the move. It does not keep several products distinct. Count the row on the still, count it again on the clip, and split into single-SKU clips when the row fails.
Ecommerce teams, marketplace sellers, and agencies use this when a collection page, a range ad, or a family shot needs two or three real products in one frame. Founders with a short catalog use it when the small size and the standard size have to stay different. After you work through the page you should be able to write a left, center, and right identity map, run a count check, walk a two-bottle plan, and stop a failing row before it reaches a listing.
Cliprise can help you approve that still and test short motion from it. Start with an image-to-video pass only after the still has a written count. If you still need the picture, build or clean it with the AI image generator, or composite real product photos, then come back to the map. The web app is at cliprise.app.
Why a standing row breaks
A single hero can warp a little and still be the same product. A lineup fails in a stricter way. The shopper can no longer tell the SKUs apart, or the frame contains a product you do not sell.
Watch for three outcomes before you comment on lighting.
Merge. Two bodies become one wider body. The gap closes, caps fuse, or a label stretches across both. Smooth motion can still be a wrong row.
Size swap. The short SKU grows, the tall SKU shrinks, or both drift toward a middle height. The row starts to read as copies of one size. A clean edge does not prove the size is right. Scale is its own review, and the method lives in AI video product scale.
Invented product. A third bottle, a new color, an extra cap, or a labeled reflection appears in the gap or at the frame edge. Two colorways can also sprout a blend color between them. That blend is a SKU you did not approve, even if nobody on the team meant to ask for it.
A same-family row, such as two sizes of one bottle, tends to average into a middle height or to copy the taller product into both slots. A contrasting row, with different caps or body colors, tends to invent a blend between them. Record the cue that actually separates the SKUs. If the only difference is a tiny word, set that word in the edit and keep a visible difference in height, color, or label shape.
A label that jumps bodies, two finishes that become one sheen, a cropped bottle completed into a new product, or a hand between the products belongs on the same reject list. No model choice removes the count.
| What you see | Name it | Next action |
|---|---|---|
| Two outlines become one, or the gap closes | Merge | Stop the lineup and split |
| Heights move toward each other | Size swap | Stop the lineup and split |
| A new body, cap, color, or labeled reflection | Invented product | Reject the clip, and reject the still if the extra product is already in frame one |
| A label cue sits on the wrong body | Label swap | Stop the lineup and split |
| Count holds, but the camera move smears the row | Motion too large | One locked retest on the same still, then split if identity breaks |
| Count holds, but a slot changed color or sheen | Color or finish miss | Fix that on its own check, and do not call the lineup passed |
Use a standing lineup for this job only
Use a standing lineup when all of these are true:
- Each item is a finished SKU you sell, or a finished component of a bundle the buyer is supposed to recognize.
- Two or three products fit fully in the frame with a visible gap.
- The difference is visible in height, color, closure, or a broad label shape. Tiny type is not the only difference.
- The shot shows a range, a family, or a collection. It shows what exists.
- You have a still that already passes the count, or you can build one before any motion.
Leave this workflow when the job is different:
- The camera should look down on a kit. That is the flat lay workflow.
- You have one hero. Animate that photo with product photo to AI video.
- The pack is still a concept and you do not have finished goods. A concept row is not a lineup of products a customer can receive.
- You need the shopper to read small legal text, a tiny scent name, or a claim. Follow logo and packaging text preservation and set exact words in post.
- The assignment is to argue that one option differs from another. Plan that in the comparison ads workflow.
- The product is mainly glass, a fill line, or a pour. Settle that accuracy problem before you ask a row of transparent packs to hold still.
- A restrained test already merged, resized, or invented a product. Split. Another adjective will not become a control.
Three angles of one bottle are not a lineup. If one SKU is copied into every slot, the clip fails even when the body count matches. A bundle can use a standing row when each component is a finished piece and the page says the buyer receives the set. An extra piece is an invented product.
Write the identity map first
The identity map names who stands where before anyone generates. Use frame-left, center, and frame-right. Avoid "first" and "second." A small camera move changes which product feels first, and a script read in another language can reverse a verbal order. Frame-left stays frame-left on the sheet, in the prompt, and in the review.
| Slot | What to record | Why it matters |
|---|---|---|
| Left | SKU id, height class, color, finish, closure, label cue | Stops the left product from borrowing traits from the right |
| Center | A real middle SKU, or the word empty | An empty center makes a third product obvious |
| Right | SKU id, height class, color, finish, closure, label cue | Stops size and label drift toward the left |
For each occupied slot, add the catalog id, height against the neighbors (shorter, same, or taller), color and finish in plain words, the closure, a label shape rather than a paragraph of claims, and one forbidden change. Copy a printed volume onto the sheet as a human note. Do not expect the model to measure.
When the row has two products, the center is empty surface: no body, no cap, no label, and no contact shadow of its own. Leave a gap you can see, about half the width of the smaller product, so a merge is obvious. That width is art direction, not a control that forces the model to comply. When the row has three products, the center is a full SKU. Prove two, then three. A longer shelf adds neighbors to blend. Stop adding bodies when the count fails.
Note anything that could be counted as another SKU. A new labeled reflection, a body cut by the frame, a spare cap, or a dark fold in the background can be redrawn into a product. If the still has no reflection, the clip should not grow one. If a teammate cannot fill the center cell with a SKU id or the word empty, the shot is not ready. Keep the map beside the still even if the tool shortens the prompt.
Run the count check
The count check is a set of numbers. Do it on the still, then on the clip. A confident glance is not a count.
On the approved still, count:
- Standing bodies.
- Closures you can point at.
- Front labels you can point at.
- Distinct colorways you intend to sell in this frame.
- Contact shadows on the ground plane.
- Bodies cut by the frame edge. This number should be zero.
- Products in the center gap. This number should be zero when the center is empty.
Write the numbers on the map. "Looks like two bottles" is how a third cap survives into motion.
On the clip, use the same numbers at the first frame, a middle frame, and the last frame, then scrub the gap and the cap line. A short failure still fails the clip. If every test needs a bad moment cut out, split.
Reject a changed body count, a shared outline, heights that move together, a new closure color, a new shape in the gap or at the edge, or a label cue on the wrong body. Write the result in those words, such as "middle frame, left bottle grew, size swap."
Flicker and a harsh grade are later notes. Do not let a lighting comment hide a wrong count. A pass on identity is not approval of label text, color, or a marketplace file. If frame one already matches a still that contains an extra product, fix the still before you spend video credits.
Prepare the standing still
Motion will not repair a still that already merged two products or cropped a cap. Fix the picture, then animate.
Photograph the finished products together, or composite real cutouts, on one ground plane. You set the gap, the heights, and the light. A generated still can explore a range before a shoot, and it still needs the map and the count. Generating the still does not make the products distinct. Compare it with the real SKUs before you spend video credits.
- Put down one ground line and stand every base on it.
- Place products at the scale you sell. The tall SKU stays taller. Do not stretch each product to the same height to balance the frame.
- Leave a gap you can see at the size the ad will run. Bases that touch hide a merge.
- Match light direction. Highlights from opposite sides get redrawn, and the redraw can change a silhouette.
- Keep only the contact shadows that belong to those bodies.
- Fill the map from the pixels, then count. If the count fails, correct the still before any motion.
The frame should show full bodies, one artwork version, a simple background, and no hands or price tags. View it at phone width. If you cannot point to each SKU, it is not ready for a tight crop.
Prove a single product with a quiet move in product photo to AI video before you stand it beside a neighbor. Check the swatch with color consistency and the surface with material finish. The lineup question is whether each slot kept its own cue.
Choose motion that does not invite a blend
The lineup test is a preservation test. Start with a locked camera and a very slow light shift. If that count passes, try one very small push that keeps every product fully in frame. Stop there. A larger move belongs on a single-SKU clip or in an edit you control.
Leave orbits, morphs, size changes, clinks, pours, spins, entering hands, and fast parallax out of the identity test. Those moves change which product sits on the left, or how tall it looks.
Change one variable per test. If you change the model, the prompt, and the still together, you cannot tell what caused a pass or a fail.
Prompt formula:
Approved still of [count] finished products. Left: [height, color, closure, label cue]. Center: [SKU or empty gap]. Right: [height, color, closure, label cue]. Keep each height, color, closure, and gap. Do not add a product. Do not merge the products. Do not make them the same height. [One motion.] Products stay planted. No hands. No new text. No props.
Those lines are instructions, not a lock, and they do not mean the model will keep several products distinct. Repeat them in any avoid field, and keep them in the prompt you log. Leave prices and competitor names out. Add words in the edit after the count passes. Use the shortest duration that still shows a first, middle, and last frame. Prefer 16:9 or 1:1. A 9:16 crop usually cuts the outer SKUs, so a vertical ad should be single-SKU clips in sequence.
Worked example: two bottles (illustrative)
The names, volumes, colors, and prompt in this section are illustrative. This is a planning walkthrough. It is not a recorded generation, not a real catalog, and not evidence that any model kept both bottles distinct.
Picture two finished bottles from a fictional range called Harbor.
- Frame-left: Harbor Travel, 100 ml, short matte sage bottle, white disc cap, small cream label with a narrow sage band.
- Center: empty warm-gray surface. The gap is about half the width of the Travel bottle. Nothing stands there.
- Frame-right: Harbor Standard, 400 ml, tall matte sage bottle, white disc cap, larger cream label with a wider sage band.
They share a body color and a cap color on purpose, so height and label size have to carry identity. This is the row that most often averages into one medium bottle or two tall copies.
| Slot | SKU | Height | Color and finish | Closure | Label cue | Forbidden change |
|---|---|---|---|---|---|---|
| Left | Harbor Travel, 100 ml | Short | Matte sage | White disc cap | Small cream label, narrow sage band | Must not grow |
| Center | Empty | None | Warm-gray surface | None | None | Must not gain a bottle, cap, or shadow |
| Right | Harbor Standard, 400 ml | Tall | Matte sage | White disc cap | Larger cream label, wider sage band | Must not shrink |
Count before any prompt: 2 bodies, 2 caps, 2 labels, 2 contact shadows, 0 products in the gap, 0 cropped bodies, 0 product-like reflections. Both bases share one line on a plain cyclorama, with no props, hands, prices, or extra words.
Illustrative prompt:
Approved still of two finished bottles. Left: the short matte sage bottle with the white disc cap and the small label stays short. Right: the tall matte sage bottle with the white disc cap and the larger label stays tall. The gap stays empty warm-gray surface. Keep both heights, both caps, and both label sizes in their own slots. Do not add a third bottle. Do not merge the bottles. Do not make them the same height. Locked camera. Very slow light shift only. Products stay planted. No hands, no pour, no new text, no props.
Judge the frames as checks, not as a finished case study. A merge is a closing gap or one cap across both bodies. A size swap is Travel growing or Standard shrinking until the heights agree. An invented product is a third bottle, cap, shadow, or labeled reflection in the gap. A label swap is the narrow band on the tall bottle, or the wide band on the short one.
A correct first frame does not save a later break. A scrub that shows the gap collapse and then reopen is still a fail.
When any of those appear, split the same day. Build two stills at the same camera height, ground line, light, and amount of floor: Harbor Travel alone, and Harbor Standard alone. Do not enlarge Travel until it fills the frame Standard occupied. Animate each with the same locked light shift. In the edit, cut from one to the other or place them side by side with a gap you set. Do not generate a clip that fuses those two outputs into one row.
Log it before the next run, with no credit amount invented: still harbor-lineup-still-v1, left Harbor Travel 100 ml, center empty, right Harbor Standard 400 ml, count 2, variable locked light shift, result named after you watch, next action split.
Replace every cell with your real SKUs before you generate. If the pair differs by cap color more than by height, judge the cap. If a third colorway must stand in the center, name that SKU and forbid a mix of the other two. A new color in the gap is an invented product. That is a map rule, not a test result.
Split into single-SKU clips when the row fails
If the standing row merges products, swaps sizes, moves a label onto the wrong product, or invents a product that is not on the identity map, stop generating the lineup. Make one clip per SKU from its own still. Assemble the range in the edit only when the timeline still needs more than one product.
Match the set, the camera height, the ground line, and the light. A small product seen from above next to a large product seen at eye level looks like a size swap you built yourself. Keep the tall SKU taller in the frame. If a lone product starts to look heroic or tiny, check product scale on that still before you animate it. Use the same quiet move on each clip, and change one variable at a time.
Cut from one SKU to the next, or place the clips side by side. You choose the gap, the order, and any names, set as real type after the pictures pass. Do not run image-to-video in order to blend those outputs back into one generated row.
One quieter retest is reasonable only when the first frame was correct and the move was too large, such as an orbit you have not yet replaced with a locked camera. If that locked test still breaks identity, split.
Deliver the passed single-SKU clips, plus an edit if you need a sequence. Name the failed row reject-merge or reject-size-swap. A collection page can use a wide or square family frame. A single-SKU page should lead with that SKU. Count a 9:16 crop on its own, and check the marketplace rules you sell under before upload.
Keep scale, color, finish, and type on their own checks
The count can pass while the product is still wrong.
Scale. Compare cap tops and bases with the still. The taller SKU is the reference for the shorter one. Drift toward one height is a size swap, and a bigger camera move will not repair it. Product scale covers believable size beside a known reference.
Color. Each slot has its own swatch. One grade on the whole frame can pull both labels toward a middle hue, which invents a colorway without adding a body. Judge left and right separately. Follow color consistency.
Finish. Matte and gloss should not share one highlight pattern. A metal cap should not turn into plastic, and it should not spread onto the next cap. Use material finish when the surface, not the count, is the question.
Labels. A lineup is a weak place to prove spelling, badges, or barcodes. The label cue is a shape in a slot. If a word must be exact, composite it. Follow logo and packaging text preservation. When two SKUs differ only by a small word, set that word in the edit. Splitting the row removes the merge. It does not approve a warped label on the single bottle.
A range shot is not a comparison ad
A standing lineup of your own finished SKUs says those products exist. A comparison ad says one option differs from another, and the two sides have to be shown fairly.
| Job | What the frame is for | Where to plan it |
|---|---|---|
| Range or family | Show finished SKUs you sell, side by side | This page |
| Overhead kit | Show items laid out on a surface | Flat lay workflow |
| One product | Animate a single hero | Product photo to AI video |
| Comparison | Argue a difference with matched, fair framing | Comparison ads workflow |
Do not invent a worse product, a competitor pack you have no right to show, or a price on the bottle. If a price must appear, set it in the edit from the current offer. Nothing on this page states a price or a credit cost. When the job is the argument, use the comparison ads workflow. Split clips can be the pictures. The claim and the fairness check stay there.
Test models by how they fail
Pick candidates from the current models list. Use the same approved still, the same prompt, and the shortest duration that shows a start, a middle, and an end. Change only the model between those runs.
Score each output as merge, size swap, invented product, label swap, or pass. A sharper clip that adds a bottle is a reject. Do not describe a model as one that keeps multiple products distinct. A pass belongs to that still and that prompt. If locked tests fail, split.
Read the credits shown for the run and check current pricing before a batch. Approve the still first. No number here is a price. Cliprise can hold the stills in the AI image generator, the motion in the image to video AI generator, and the current list on models. That does not make a row stable. Log a seed if one is shown. It applies to that still only.
Mistakes that send the wrong product out
- Starting from a sentence and no still you can count.
- Treating an overhead flat lay as a standing lineup.
- Letting the products touch so a merge hides in the contact.
- Cropping a cap or a base and expecting the frame to finish it honestly.
- Using an orbit as the first test.
- Asking the small SKU to become the large SKU.
- Copying one lucky prompt onto a longer shelf. Prove two products, then three, and stop when the count fails.
- Treating a pass as a property of the model.
- Writing claims, prices, or badges into the prompt.
- Uploading the range clip as the only media on a single-SKU page.
- Letting a reflection become a third labeled product.
- Regenerating an identity fail instead of splitting, or saving that reject as final-v3.
The count is how you catch the wrong size, or a product you do not sell, while the file is still a test.
Log the lineup so the next test is real
A lineup log follows an AI video generation log, with slots added so the next person can see which product moved.
| Field | What you write |
|---|---|
| Still id | Filename of the lineup still or the single-SKU still |
| Left, center, right | SKU id, or empty |
| Expected count | Bodies, closures, labels, shadows |
| Model route | The model you actually selected |
| One variable | The only thing this run changed |
| Duration and aspect | What you requested |
| Result | Pass, merge, size swap, invented product, label swap, or other |
| Credits shown | The figure the product displayed for that run |
| Next action | One quieter retest, or split |
Keep the still with the entry. Log the lineup failure, then each single-SKU clip on its own row. A pass should say you checked the first frame, a middle frame, the last frame, and a scrub.
Checklist before the clip leaves the team
- The map names frame-left, center, and frame-right. Center says empty or names a real SKU.
- The still shows finished products, a visible gap, and no cropped bodies.
- The count on the still is written down.
- The prompt has one motion and the forbidden changes.
- The first frame, a middle frame, and the last frame match that count.
- A slow scrub shows no collapsed gap and no drifting cap line.
- Color, finish, and the label cue still belong to the correct slot.
- No invented price, claim, or competitor package is in the frames.
- If identity failed, the exports are single-SKU clips and the row file is named as a reject.
- A collection can use a passed lineup or an edit. A single-SKU page does not use the multi-SKU clip as its only picture.
Next step
Pick two finished SKUs. Write the left cell and the right cell, and mark the center empty. Approve a standing still you can count. Run one locked test in the image to video AI generator. If the count holds on the first frame, the middle, and the last frame, review color, finish, and type before anyone publishes. If the count breaks, split that same day and animate each product alone.
Record the run in an AI video generation log. Leave overhead kits on the flat lay page, a single hero on product photo to AI video, and an argued difference on the comparison ads workflow.
The lineup is finished only when a person can name every product in the frame and can say that no extra product appeared.