Patient monitors: bezel still; waveforms are overlay, not Veo vitals
Photograph the bezel. Overlay traces you actually ship. Veo will invent numbers. No talking ICU twin.
Photograph the bezel. Overlay traces you actually ship. Veo will invent numbers. No talking ICU twin.
A patient-monitor listing is a chassis a biomedical engineer already specified: this bezel, this alarm strip, this connector row. Heart rate, SpO2, and the ECG sweep are strings you already print. Asking Veo 3.1 to “show live vitals” mints a 72 that was never on the SKU, and a talking white coat next to that glass is a constructed clinician. Lock the bezel. Typeset the traces. Leave the ICU off the generate.
Photograph the bezel you ship
The buyer is matching a serial to a drawing. Bezel colour, alarm LED, the badge, the port block — those are identity. Text-to-video asked for “a sleek ICU monitor on a busy ward” invents a cousin chassis. The applications engineer who pauses on the bezel notices before they notice your logo.
Photograph the unit you catalog. Empty glass or a dark demo screen with no readable numbers. Crop people. Crop a live waveform if a real case was on the glass. A generated patient or extra in scrubs is a hard stop. The chassis is the listing. The ward is not.
Image-to-video starts from that frame. Prompt camera and physics: a slow orbit, a hold on the ports. Re-describing “a clinical-grade monitor in a cinematic ICU” draws a second chassis. If the physical badge is already in the still and it is this unit, keep it. Do not “sharpen the stamp” inside the generate.
GPT Image 1.5 is chassis, not a vital
If you already have a catalog still of the bezel with dark glass, keep it. Do not redraw a chassis you can shoot. Generated type is for the graphic that is not in the file yet — a lockup or end card you will still overlay live numbers on.
GPT Image 1.5 is OpenAI’s row across text-to-image, image-to-image, and edits. It is strong at long, specific instructions and legible in-image text. Content type is image. Max output is 4K. Display price is 2 credits. Audio is null. Aspects include 1:1, 2:3, and 3:2.
Quote strings that belong on a product plate, not a live case: SKU, series name, “demo screen dark.” Proof at 100%. Do not ask this row to invent an ECG or a 98% SpO2. Those glyphs are claims. Claims you will ship belong in an editor you can correct without regenerating the bezel. Attach the catalog jpeg and describe only the plate change if the iron is already right. Re-describing “a busy ICU bay” draws a second room. Fail a badge audit: throw the still away.
Overlay traces you actually ship
Muted autoplay is how these clips get watched. The numbers belong on screen. Speech is optional. The overlay is not.
Type the line from the demo mode or the sheet you actually publish. Heart rate, then SpO2, then a “see the IFU” beat. A model asked to paint those onto Veo glass writes something adjacent: a plausible 80, a cousin sweep. Keep the wording in an editor.
Add timed text overlays to a video burns authored lines at start_sec and end_sec. Font, size, colour, and position apply to the set. Nothing is transcribed. A mute bezel has nothing to listen to. If you only need one persistent badge — one SKU, one “demo only” line — that is add a text overlay: one string, whole duration, top / center / bottom.
Crop to the ship ratio before you overlay. Picture lock, then type. Demo numbers change when the sheet changes. The bezel should not. Do not prompt Veo to letter the glass. A 2 that becomes a 7 for six frames is a mis-ship. Waveforms are typesetting, not physics.
Veo invents numbers; a talking ICU twin is a kill
Veo 3.1 is native audio, official 4K tier, 4 / 6 / 8 second clips, 40 credits. That is a cinematic talking row. A monitor loop does not want that mouth. “Professional intensivist, here’s the SpO2 you should watch” spends the talking meter on a fabricated rating and a presenter. Overlay after picture lock. Keep lips out of frame.
YouTube’s third inauthentic-content bucket is AI-generated personas that present as human experts on health, legal issues, finances, or politics. A constructed ICU clinician walking a protocol or “what this waveform means” is the same shape as a constructed doctor diagnosing a rash. Labelling the face AI does not leave the bucket.
A silent bezel is not that shape. Authored type is typesetting, not a mouth. A real, named, credentialed employee on camera is a different day — and they still should not typeset vitals while a model redraws the glass. Do not glue a talking white coat onto the bezel file. Do not generate a patient so the monitor “looks in use.” Empty glass. Overlay the demo line you actually ship.
FAQ
Can Veo 3.1 typeset HR and SpO2 if I write the numbers in quotes?
Do not. Prompted type in a video generate is a miss you cannot edit, and Veo will still invent a chassis around it. Photograph the bezel. Overlay the traces you actually ship after picture lock. Veo 3.1 is the 4 / 6 / 8 second talking and 4K row, not a typesetter.
Do I still need GPT Image 1.5 if the catalog photo already has dark glass?
No. Keep the photo. GPT Image 1.5 is for type that is not in the file yet, or for an edit on a locked still. Animate the real bezel with image-to-video. Do not pay a lettering model to redraw a chassis you already shot, and do not ask it to invent a live ECG.
One overlay or timed traces?
One persistent SKU or “demo only” badge is a single overlay on the whole file. Heart rate, then SpO2, then a sheet URL is timed overlays, with timestamps you set. Neither job transcribes. Neither job belongs in the Veo prompt.
Can we generate a photoreal ICU nurse so the monitor looks in use?
No. A photoreal nurse is a constructed health presenter. A photoreal body on a bed is a generated patient. Photograph the empty bezel. Overlay the demo traces. If someone must speak, that someone is a real named credentialed person on a different shoot, not a generate.