Bento: overlay the compartment names; do not invent the food
Animate the box still and burn the compartment labels; generated fillings are a different SKU.
The box still is I2V. Labels are burned. Generated fillings are a different SKU.
A bento brand sells the tray, the lid, the gasket, the wells. Motion belongs on a photograph of that box. Compartment names — rice, protein, veg, sauce — are strings you typed. They are not a generate of salmon that is not in the well. If the still is empty, keep it empty. If the still is a real packed meal you photographed, keep those fillings. Do not ask a model to invent lunch.
The box is the product. The filling is a claim.
Open the lid in the photo you already shoot for the PDP. Same plastic, same divider height, same latch colour. That file is the brief. Image-to-video moves it: a slow push, a lid lift you actually framed, steam only if steam was in the still. The AI image-to-video tool is that door. No still, no generate.
A packed box you did not cook is a different SKU. Generated rice sits at the wrong grain. Generated protein sits in the wrong well. A buyer who owns the tray will see it in one pause. Catalog honesty is the empty or the real pack — not a cousin bento the sampler plated.
Do not text-to-video "beautiful Japanese lunch box with salmon and avocado." That sentence rebuilds the tray. You already own the tray. Attach it.
/for/restaurants is a kitchen and a plate as lifestyle. /for/personal-chefs is a cook in someone else's home. This page is the box. Do not rewrite those audiences as a compartment layout, and do not import a restaurant's invented garnish onto an OEM still.
Motion on the locked tray, then type
Pick an image-to-video row that will honour the still. Wan Video 2.5 Image to Video is a cheap silent pass when the plate is already the product: requires an image, audio false, 5s or 10s, up to 1080p, 15 credits. Medium motion on an approved lid is often the whole job. High motion on an unapproved pack is ten seconds of drift and a new filling.
Prompt the camera, not the menu. "Slow dolly across the closed lid, hold, no new objects." Stop. "Add spicy tuna to the left well" is a different generate. It will not match the SKU. It will not match last week's overlay. It will not match the meal the warehouse packed.
Silent is useful. Muted autoplay still needs the names. That is overlay, not Wan glyphs. Do not ask the motion model to typeset "protein" on the lid. Painted type is a miss you cannot edit when the merchandising line changes.
Overlay burns the names you wrote
Add a text overlay to a video stamps one authored line on the whole file — top, center, or bottom. It does not listen. It does not transcribe steam. "Rice · protein · veg" is a string you typed. Burn it after picture lock and after the crop. A 16:9 overlay that then gets reframed to 9:16 sits in the chrome.
One line, whole clip. If the names need to change at 0s, 3s, and 6s, that is timed overlays, not this persistent headline, and not auto-captions of a silent tray. A text overlay burns one authored line; it does not invent the food under it.
Do not generate the labels in the still. Compartment print on a lid is a brand asset you already have, or it is overlay. A model that "helps" with the word rice will help you into a typo. Keep the plastic clean. Print after.
What this is not
It is not a chef talking over macros. It is not a Veo packed lunch from a paragraph. It is not Omni B-roll of a Tokyo kitchen that happens to include a box. It is not a first-last of empty-to-full unless you photographed both states of the same tray.
Empty-to-full is a different contract: frame A the empty wells, frame B the real packed meal, same lid geometry. Interpolation can travel that. A single still plus "fill the compartments" is not that contract. It is a new SKU.
If you sell inserts — ice pack, sauce jar, utensil slot — shoot those pieces. Animate those stills. Overlay the insert name. Do not prompt a jar into a well that does not take one.
The job on this page is small: lock the box, move the box, burn the names. Food you did not photograph stays off the generate.
FAQ
Can I prompt the model to add salmon to the protein well?
No. That filling is a different SKU. Photograph a real pack if you need food in the wells. Image-to-video moves what is already there.
Should the compartment names live in the generate?
No. Overlay them. Generated type on a lid is a misspelled merchandising line you cannot edit when the names change.
Do I have to shoot a packed box?
Only when the ad is the meal. The tray SKU can be empty. Empty is honest. A generated pack is not.
Why not restaurant B-roll of someone eating from the bento?
That is a restaurant or chef still, and it is talent. This page is the box. Lifestyle eating is a different shoot, with a real person, not a generated diner.