Puzzles: first-last empty table to finished image if you own both stills

Two photos.

An empty table. A finished image. If you own both stills, the clip is interpolation: open on the blank, land on the assembled puzzle. If you only own one, you do not have a first-last job yet. Generating the missing still is how a licensed image becomes a cousin of itself, and how a SKU becomes a different box.

Create a first-and-last-frame transition is the named job: two photos in, one clip that opens on the first and finishes on the second. First-last frame is the control. A paragraph about "cinematic time-lapse of a puzzle assembling" is not.

Empty table and finished image are two files

The crew already photographed the blank. They already photographed the completed puzzle — or the box art you have the right to show as the destination. Those two files are the brief. Same table, same lens height, same standing spot. A phone still of a bare wood top and a scan of the box from a different afternoon is not a transition. It is a warp.

Attach both. The agent passes them as first and last frame and picks a first-last-capable model. Output is a single clip that travels from empty to finished. Cost is priced per model, shown before you confirm.

Ordinary image-to-video knows where it begins and improvises the rest. Give it only the empty table and it will invent a picture in the middle. That invented picture is not your SKU. If you never attach the finished still, there is nothing to land on. A macro of one piece to a wide of the whole table in two seconds is a warp, not an assemble. Lock the tripod or lock your feet. Then interpolate.

Interpolation is not a product restage

The AI product video generator is a different door. Reference-to-video holds a real object's shape, colour, and packaging while the scene changes. Two or three clean angles are the contract. Fine print is the first thing to inspect at 100%.

Do not mix the contracts on one take. Empty-to-finished is first-last: both ends pinned, motion in the middle. A product hero in a new room is R2V: identity travels, composition does not. Feeding the finished puzzle still to R2V and asking it to "just assemble" will restage.

First-last frame is the product-turn model is the same contract applied to a SKU. The assembled puzzle is the SKU. The empty table is frame A. Do not prompt a text-to-video row to "morph" and hope it honours a second still you never attached.

If you do not own the finished image — a licensed illustration, someone else's photograph, box art you cannot use as a destination — wait. A generated "after" is a different puzzle.

The middle is where a generator cheats

Once both stills are attached, the prompt is only travel. Slow locked-off melt. No handheld. No new table. No extra pieces that were not in either photo. The stills are the spec. Language does not pin pixels.

Duration should match the morph, not a Reels quota. A short travel between two honest table photos reads as proof. A long travel invents a second table and a picture that is neither still. Inspect the midpoint. Ends are held by construction. If the middle grows a phantom piece, reroll the interpolation, not the stills.

The test: do I know the start picture and the end picture, and is the point to travel between them without inventing a new image?

Yes → first-last. Attach empty. Attach finished. Generate once.

I only have the empty table, and the end can wander → that is turn a photo into a video, and you should not publish it as the assembled SKU.

FAQ

Can I first-last a puzzle if I only photographed the empty table?

No. That is a hold of the before, plus an invented after. Empty-to-finished needs both stills. Shoot the assembled image, or wait until you own it.

Do both photos have to be the same table?

Yes. Same site, same standing spot, same lens family. Unrelated plates will connect, and the connection is the artefact everyone notices. The agent job will not rescue a mismatched pair.

Is the product-video generator the same as first-last?

No. Product video is reference-to-video: the box stays yours while the scene changes. First-last pins two frames you already own and interpolates. Empty table to finished image is the second job.

What if I want a 30-second identity pass of the finished puzzle?

That can be a still-to-motion pass of frame B, after the interpolation exists. It is not a substitute for the interpolation, and it still starts from a still you own.