VLM OEM: bay still, overlay tray count, not cinematic forklifts
Pallet racking already owns bay geometry. The VLM listing is the tower still. Overlay the tray count you sell. Veo will invent a different carousel.
Pallet racking already owns bay geometry. The VLM listing is the tower still. Overlay the tray count you sell. Veo will invent a different carousel.
A vertical lift module is an enclosed cabinet: extractor, tray face, access opening, the paint you ship. It is not a selective-rack aisle with a taller crop. Beam class already has a home. This listing is tray count on a tower a buyer can match to a drawing.
The tower you sell is not a racking bay
Pallet racking: bay-geometry still, overlay beam rating, not cinematic forklifts already named the open-iron job: upright pitch, beam length, connector, a plate a safety walk will pause on. Do not rewrite that post as a carousel. A VLM does not take a pallet the way a bay does. The operator works an opening. The extractor travels. Tray pitch is the geometry.
Photograph the cabinet you catalog — square-on or a three-quarter that still reads the door, the access window, the tray in presentation if that is the line-card pose. Same lens height as the print. Empty enough that tray edges count. No generated picker. No invented reach truck feeding a machine that is not an aisle.
Text-to-video has no tray. “Cinematic warehouse, forklift staging a vertical carousel” will mint a cousin cabinet, a smear for a count, and extras you did not hire. Average is not your pitch. Average is a different SKU.
Overlay the tray count you sell
Muted autoplay is how these clips get watched. Tray count belongs on screen. Speech in the take is optional. The overlay is not.
Type the number from the data sheet you actually publish — tray quantity, payload per tray, the footprint string if you print it. A model asked to paint that onto a generated door will eventually write something adjacent. Keep the wording in an editor you can correct without regenerating the iron.
Add timed text overlays to a video burns several authored lines, each with start_sec / end_sec. Count, then payload, then a “see the line card” beat. Font, size, colour, and position apply to the set, not per line. Nothing is transcribed. A mute tower has nothing to listen to. If you only need one persistent badge for the whole clip, that is add a text overlay: one string, whole duration, top / center / bottom.
Crop to the ratio you will ship before you overlay. Picture lock, then type. Counts change when the line card changes. The cabinet should not. Overlay the new number. Do not regenerate the tower to change the plate. Do not ask the generate to letter “24 trays” into the fascia. A 2 that becomes a 7 for six frames is a mis-ship.
GPT Image 1.5 only if the tray graphic must be drawn
If the count is already on the fascia in a photo you took, keep the photograph. Do not redraw a plate you can shoot. Generated type is for the graphic that is not in the file yet — a lockup, an end card, a tray-face render you will still overlay the live number on.
GPT Image 1.5 is the catalog row when you must render type: text-to-image, image-to-image, and edits; 4K; 2 credits; strong at long, specific instructions and legible in-image text. Quote the string. Case, alignment, how many lines. Proof at 100%. A wrong digit is a different machine.
Use it as a still job. A tray count, a short payload line, a SKU stamp. Do not ask it for a full code block in one prompt. Edit is on the same slug: attach the catalog jpeg and describe only the plate change if the cabinet is already right. Re-describing “a cinematic VLM aisle with a forklift” is how the model starts drawing a second tower. Fail a plate audit: throw the still away. Do not animate a misspelled count.
Veo will invent a different carousel
Veo 3.1 is native audio, official 4K tier, 4 / 6 / 8 second clips, 40 credits. That is a cinematic talking row. A VLM tower does not want that mouth or that extra. “Professional warehouse, reach truck, tray count on the door” is how you spend the talking meter on a forklift that is not yours and an extractor that is not the drawing.
Cinematic forklifts are extras. Generated operators are people you did not hire. Empty-enough tower. Overlay the count. Leave the truck out of the generate unless it is in a still you actually shot of your floor — and even then it is not the product. The product is the cabinet.
If the still must move, image-to-video starts from the approved tower. Prompt camera and extractor travel only: a slow hold on the access opening, a tray presenting, no new slots. Re-describing the warehouse is how the model draws a second carousel. Do not morph tray pitch. Do not grow an open bay on an enclosed unit. If the generate redraws those, it is a mis-ship.
Photograph the tower — or GPT Image 1.5 only for plate letters you proofed. Then hold: lock the still, move the camera, burn the tray count you sell. Do not Veo a forklift so the aisle looks busy.
FAQ
Can Veo 3.1 typeset the tray count if I write it in the prompt?
Do not. Prompted type in a video generate is a miss you cannot edit, and Veo will still invent a carousel around it. Photograph the fascia as a still. Overlay the number you actually sell after picture lock.
Do I still need GPT Image 1.5 if the count is already in the photo?
No. Keep the photo. GPT Image 1.5 is for type that is not in the file yet. Animate the real tower with image-to-video. Do not pay a lettering model to redraw a plate you already shot.
Why will a video model invent a different carousel?
Because text-to-video has no plate. You asked for “cinematic warehouse with a forklift,” and the model has to decide whose extractor. Average is not your tray pitch. Average is a cousin cabinet and a smear for a count.
Is this the pallet-racking post with a taller crop?
No. Pallet racking is bay geometry plus a beam rating. This page is an enclosed VLM tower plus a tray count. Different iron. Different overlay. Do not copy a beam-class prompt onto a carousel.