POS hardware OEM vs /for restaurants: the terminal still is the SKU
/for/restaurants is the room; the OEM listing is Seedance on the terminal still, with screen type in Ideogram or overlay, not Veo UI.
/for/restaurants is the room. The OEM listing is Seedance on the terminal still. Screen type is Ideogram or overlay, not Veo UI.
A restaurant fills a table with the dish. A POS OEM sells a catalog cut: this chassis, this bezel, this SKU plate. Mixing them in one generate is how a factory still becomes a fake dining room, and how a plating reel parks a cousin terminal on the counter. Photograph the unit. Animate the unit. Letter the UI somewhere you can edit.
/for/restaurants is the room, not the chassis
That audience page is a kitchen and a table. The winning hook is the dish, not the logo: steam off the pass, a cheese pull, a plate landing. Five posts a week. 9:16. The food is the product.
If you sell terminals, that page is the wrong brief. Do not copy a cheese-pull calendar onto a PIN pad. The plate is the restaurant's product. The bezel is yours. Do not rewrite /for. Name the difference and stop.
Nearby cafes sell morning ritual — a pour, a window seat. That is still a room. Keep the terminal large. Crop the dining room before you buy seconds. A generated host walking "tonight's specials" is a presenter. The OEM listing needs neither.
Seedance on the terminal still
The buyer on the factory side is matching a serial to a drawing. Bezel colour, card-slot geometry, the badge, the stand — those are identity. Text-to-video asked for "a sleek POS on a sunlit bistro counter" will invent a dining room and a cousin box. The distributor who stocks this SKU will see the miss.
Image-to-video starts from the frame you upload. Frame one is the file. The prompt is camera and physics: a slow orbit of the chassis, a tilt to the ports, a hold on the plate. Re-describing the terminal in the prompt is how the model starts drawing a second chassis.
Seedance 2.5 is the still-to-motion family: four to thirty seconds, native audio, 720p, 47 credits listed. I2V is the still-to-motion job on that family; the published URL is this slug. Thirty seconds in one pass is why this family holds a SKU. Stitching four short text-to-video clips around a bezel is how the badge drifts. 720p is the ceiling on that endpoint. If it has to remain the same terminal, you are here.
Native audio is on. Write the mix (a quiet plant floor, no speech) or you will get a soundtrack you did not choose. A last frame is optional: use one when you have a second approved still — front then ports. A single hero still with a slow orbit is the usual catalog job.
The AI product video generator is the same contract: reference photos are the ground truth. Zoom to 100% and read the small print before you keep a take. Fine text is the first thing a generator loses. If the still is a studio pack shot, keep the studio. Do not prompt a bistro into existence so the ad "feels F&B."
Screen type is Ideogram or overlay, not Veo UI
The string on the glass is a claim: SKU, firmware, a demo amount, "Tap to pay." A 2 that becomes a 7 for six frames is a mis-ship. Video models redraw letterforms every frame. A UI is typesetting. It is not cinema.
Write the video prompt as a clean plate. Dark glass, no readable ticket, no invented total. Then put the real string on later.
Ideogram V4 is the lettering still: text-to-image, 6 credits, posters and logos with accurate text listed. Quote the string you will actually ship. Proof it like a press check. Cut it as a title card, or composite it onto the glass. Do not I2V the chassis and hope the model spells the firmware on the LCD.
If the terminal still is already approved, skip a second image generate for the claim line. Add a text overlay burns one authored line for the whole clip — top, center, or bottom. That is the model string, a CTA, a contactless badge. It does not transcribe. Glyphs inside the Seedance pass are uneditable. Overlay after picture lock. Swap the string for the next SKU without regenerating the bezel.
Veo is the dining room, not the typesetter
Veo 3.1 is short cinematic video: native audio, official 4K tier, 4 / 6 / 8 second clips. That is a room, a plate landing, steam as texture. It is not a monospace renderer. Routing Veo at a POS glass is how a shop burns a premium row on a subject the overlay tool already owns.
"Professional restaurant host explaining tonight's POS" on Veo is the wrong object twice: a constructed presenter, and a UI the model will invent. Hold the chassis. A real named operator on a different day if someone must speak.
The test is who has to recognise the object. If it is a distributor matching a SKU, you are on the terminal still. If it is a diner matching a table, you are on /for/restaurants. One generate cannot be both.
FAQ
Can we use the restaurants page as a template for terminal ads?
No. That page sells the dish to fill a table tonight. The OEM job is a locked still of the chassis. Copying steam-off-the-pass film ideas onto a PIN pad is the room miss.
Why not text-to-video of a POS on a beautiful bistro counter?
Because the room is invented and the chassis will drift. Upload the unit photo and describe only the camera. Restaurant set dressing is a second, approved still — or it is not in the ad.
Can Veo 3.1 render a short UI if I put the string in quotes?
Not as a production system. Four, six, or eight seconds of native-audio cinema is the row. A quoted ticket is still in-frame type across every frame. Ideogram the lockup, or overlay it after lock.
Do we need native audio on the SKU turn?
Seedance 2.5 generates native audio with the picture. Write a quiet mix, or plan to live with the bed it invents. Spec numbers still go on as overlay. Do not expect the terminal to speak the model string.