Keep the SKU identical across shots

Keep the SKU identical by attaching one locked still to every shot that shows the pack. Run I2V or R2V from that frame. Seedance 2.0 is 69 credits, T2V+R2V, 1080p, 4–15s, native audio. Seedance 2.0 Fast Reference to Video is 25 credits, 720p, 4–15s, native audio. Do not re-prompt the bottle. A new paragraph is a new cousin.

This is identity across a cut. It is not “lock the still then I2V” as a first-product tutorial — it is how you reuse that still so shot 1 and shot 4 are the same listing.

One still is the only contract that survives a cut

Shot 1: hero pack on a table. Shot 2: talking head. Shot 3: pour. Shot 4: pack + CTA. If shots 1, 3, and 4 each T2V a “teal serum with gold cap,” you will ship three SKUs. The buyer will notice. The retailer will notice. QC should notice first.

Lock one still:

1. The listing photo or a generated still you already approved.
2. Label readable. Colourway correct. No extra flavour.
3. Crop variants if you need them (9:16 hero, 16:9 insert) from the same master, not from new generates.
4. Put the file in the brief. Every shot that shows the pack points at that file.

The talking-head shot does not need the still unless the person is holding the pack. If they hold it, the still is in their hand via I2V/R2V or a composite, not via “holding a teal bottle” in T2V.

Forbidden: a new still per shot “for variety.” Variety is camera and verb, not identity. A second still is a second SKU unless it is a documented second angle of the same photoshoot.

R2V when several shots must share the pack

Reference-to-video is the row family that takes a stack, not a sentence.

Row	Job	Duration	Resolution	Audio	Credits
Seedance 2.0	T2V + R2V	4–15s	1080p	Native	69
Seedance 2.0 Fast R2V	R2V	4–15s	720p	Native	25
Veo 3.1 Reference to Video	R2V	8s	(card)	Native	64

Pick Seedance 2.0 (69cr) when you need 1080p, native audio, and references (the pack still, maybe a location still, maybe a face still) in a 4–15s shot.

Pick Fast R2V (25cr) when 720p is enough and you are generating several pack shots in one session. Three Fast R2V pack inserts are 75 credits. Three full 2.0 shots are 207. Use 2.0 on the hero hold; Fast on the extra inserts if 720p matches the timeline.

Pick a single-image I2V row (Kling O3 Pro I2V 70cr, Wan 2.7 I2V 15cr audio-null) when one still and one verb is the whole shot. R2V earns its keep when you are stacking more than “here is the bottle.”

Procedure per pack shot:

1. Open the R2V or I2V row, not T2V.
2. Attach the same still every time. If the row takes multiple references, the pack still is always in the stack. Do not swap in a “nicer” generate from yesterday.
3. Prompt the verb and camera only: “slow push-in, label holds readable, no extra packs.” Do not describe the bottle.
4. Duration = one verb. The pour is not also the unscrew is not also the CTA.
5. Generate. Freeze-frame against the still. If the label drifted, fail the take.
6. Repeat for the next pack shot with the same still and a new verb.

T2V is allowed for shots with no pack: the face hook, the city B-roll, the hands without a label. The moment the SKU enters, return to the still.

The AI video generator will not remember the bottle from the last job unless you attach it. History is not a reference stack.

Cut the shots together without “matching in the grade”

A grade will not turn cousin bottles into one SKU. It will make them the same cousin.

On the timeline:

1. Pack shots from the same still, same row family if you can (grain, motion).
2. If you mixed Fast 720p inserts with a 1080p 2.0 hero, decide the master: downscale the hero or upscale the inserts. Do not leave a resolution pop on the CTA.
3. Watch the cut specifically for SKU continuity: cap, colour, label text, fill level. A pour shot that ends fuller than it started is an artefact, not a feature.
4. If one insert drifted, regenerate that insert from the still. Do not T2V a “fix.”
5. Only then add VO/TTS if the R2V native audio is not the line, and burn captions.

Fill level, condensation, and stacked boxes are identity too. If shot 1 is a full bottle and shot 3 is a half-empty bottle you did not pour on camera, you invented a skip. Either generate the pour as the connecting beat or keep the fill constant.

Hands: if a hand covers the label, you cannot QC the SKU. Re-prompt the verb so the label stays readable, or add a labelled insert.

Do not “fix” a drifted label with a caption that types the flavour name over the mush. The pack is still wrong. Type is not a reprint.

When one still is not enough — and when it still is

You need more than one still when:

• Front and back of the pack both appear and one photo cannot crop both.
• A true before/after (closed carton vs open) that you photographed.
• A colourway set that is actually several SKUs — treat each colour as its own still, own shots, never mixed in one R2V stack unless the ad is a lineup and you have all five photos.

You do not need a new still because the camera moved. I2V/R2V is the camera move. The still stays.

Seedance 2.5 is T2V, 47 credits, 30s, 720p. It is a duration row, not an identity row. Do not ask 2.5 to hold a SKU across 30 seconds from a paragraph. If you need 30 seconds and a pack, storyboard: 2.5 or Turbo for the face, R2V/I2V from the still for every pack second, concatenate.

If you need a studio seat to pin the still and run 2.0 or Fast R2V across the pack shots, sign up. Bring one frame. Do not bring four paragraphs about the bottle.

FAQ

Why not describe the bottle the same way in every T2V prompt?

Because the model does not have the bottle. Identical adjectives still sample a new cousin each job. The still is the identity. Attach it.

Seedance 2.0 or Fast R2V for every insert?

1080p hero and legal close-ups: 2.0 at 69 credits. Extra 720p inserts in a vertical cut: Fast R2V at 25. Same still on both.

Can I mix a T2V pack shot “just for the wide”?

No. The wide is where the cousin is most visible. I2V/R2V the wide too, or keep the pack out of that shot.