Ice trays: first-last empty grid to frozen cubes
Two stills. Interpolate. Do not generate a new tray color.
Two stills. Interpolate. Do not generate a new tray color.
An ice-cube tray brand does not need a cinematic freezer. It needs the empty grid, then the same grid full of frozen cubes. That is a bounded travel. You already own both photographs if you shot the SKU. Do not prompt a model to invent a cousin tray in a new colourway. The colour is the listing.
Empty grid is frame A. Frozen cubes are frame B.
Photograph the tray empty — same lens height, same counter, same light family. That is the first frame. Photograph the same tray frozen, cubes seated in the same cavities, same standing spot. That is the last frame.
First-last frame takes those two stills and generates the motion between them. Ordinary image-to-video knows where it begins and improvises the rest. First-last has to land. The arc of the shot is decided by you: empty cavities become ice. The silicone, the lid colour, the cavity count stay the SKU because they are in both photographs.
The two frames have to be plausibly connectable in the time available. Same tray, same counter, same crop. A macro of one cavity morphing into a wide of a freezer door is a warp, not a freeze. Keep the grid.
Do not treat a loop (same still in both slots) as a freeze. A loop is ambient motion on one plate. This offer needs two states.
Create a first-and-last-frame transition video is the named agent job: two photos in, one transition out. Cost is priced per model, shown before you confirm. This is not the plain "turn one photo into a video" capability. Missing the frozen still is a different job.
Veo first-last interpolates. It does not recast the tray.
VEO First Last Frame interpolates between first and last frames. That is the whole description. Image-to-video. Native audio on. Durations only 4s, 6s, and 8s. Output quality includes 720p, 1080p, and 4K. Aspect ratios 16:9, 9:16, and auto. Catalog sticker: 40 credits. It does not invent a tray from a paragraph. It does not extend a keeper. It spends those credits to travel from a still you already approved to a still you already approved.
Prompt the travel, not the object. The stills already contain the grid and the cubes. Name the mechanism: hold, then frost fills the cavities; camera locked; no new counter; tray colour unchanged. If the endpoints share geometry, the morph reads as a freeze. If they do not, you get a mushy crossfade you could have made in an editor without a video bill.
4s, 6s, 8s — there is no ninth second on this row. A freeze that reads in two seconds looks padded at eight. A four-second interpolate of a short fill is the honest length. An eight-second travel that invents a second tray in the middle is the miss. Pick the length that matches the distance between the two stills.
Native audio rides along. A silent prompt still gets a soundtrack you did not choose. Write the mix — a quiet freezer, a cube crack, no voice — or plan to replace the bed. First-last will not lip-sync a host. If a mouth is the shot, you are in the wrong row.
4K is the ceiling, not a default. Spend 4K when the plate is a hero. The sticker is 40 credits either way. Two stills that disagree about the lid colour will not be rescued by resolution.
Colour is the SKU. Do not generate a new one.
Ice-cube trays sell on colourway, cavity count, and lid. Text-to-video "blue silicone ice tray filling with cubes" recasts all three every take. Image-to-video from one empty still guesses the frozen state and often guesses a different hue. First-last pins both ends. If both photographs are the blue lid you ship, the middle has to travel through that blue.
Inspect the midpoint. Ends are held by construction. The middle is where a generator cheats — a sixth row of cavities, a lid that tints, cubes the mould never made. If the middle is wrong, reroll the interpolation. Do not regenerate both stills "to match the motion."
Do not use a last frame that is a different SKU. Pink empty into blue frozen is a morph of the product. Do not describe the colour in the prompt as a way to "fix" a still that is the wrong colourway. Fix the still. Then interpolate.
Ice cream shops is a different product
Ice cream shops sell a melt: scoop pulling out of a tub, a cone starting to drip, same-day foot traffic on 9:16. That page is texture and impulse. It is not a tray OEM listing. Copying the shop playbook onto a silicone grid — generated scoop, generated flavour drop — is how a cavity SKU becomes a dessert ad for a store you do not run.
The tray brand's job is the empty-to-frozen interpolate of this mould. Overlay a lid SKU after lock if the feed needs a line. Do not generate a new tray colour because a shop post wanted "summer." Pause on the grid. Same cavities, empty then full: this page. Real scoop in a real shop: /for.
FAQ
Can I animate one photo of the empty tray and hope it freezes?
You can. The ending is then the model's guess. Ice-cube trays are the frozen grid. Pin the frozen still as the last frame or you are not selling the freeze, and you are inviting a new colourway in the last second.
Why Veo first-last instead of a talking-head row?
Because the product is interpolation between two approved stills, not a mouth. VEO First Last Frame is 4 / 6 / 8 seconds, 720p / 1080p / 4K, native audio, 40 credits. A presenter row will talk over a tray it invented.
Do the two photos have to be shot on the same day?
They have to share geometry. Same tray, same lens family, same lighting neighbourhood. A noon still into a night still is a time-lapse you should prompt as one. A different colourway is a morph.
Is this the same as uploading one image to image-to-video?
No. One image is open-ended motion. Two images is a bounded landing. The agent notes that this capability needs first frame and last frame. Missing the frozen grid is a different job.