Otoscope specula: GPT Image 1 Edit at 2cr, 4K keeps the cone; overlay French size

GPT Image 1 Edit is 4K, 2cr — retouch the photographed speculum, do not invent a tympanic membrane. Overlay French size as authored type. No photoreal patient. /for/audiologists is the clinic page, not this SKU.

A speculum listing a buyer can match is this cone: taper, distal opening, the finish you ship, the French size already on the line card. It is not an eardrum. It is not a fitting-room conversation. Photograph the disposable or reusable speculum on a clean card. Spend two credits on the delta. Stamp the size you will stand behind. Leave the family conversation to the clinic page.

Photograph the speculum; keep the cone

The honest still is a photograph of the speculum you place. Square to the cone. Distal rim in focus. Proximal flange readable if that is how the SKU is identified. Even light. No ear. No canal. No tympanic membrane drawn in the lumen “for context.” The lumen is a hole in a plastic (or metal) cone. Empty is the catalog pose.

Text-to-image of “a pediatric otoscope exam, golden hour, smiling child” invents a patient and a cousin funnel. That is a different object, and it is a body you did not consent. Crop a handwritten lot tag you would not print. If the only photo includes a person, crop them or shoot again. A generated patient is a hard stop. Minors are a hard stop. Do not prompt a dummy canal so the cone “looks in use.”

Hearing-aid OEM is the shell still, not the audiologist /for page already spent the housing. Do not copy that page onto a speculum. A RIC listing is a receiver wire. This page is a cone.

Do not ask a video model to invent the funnel. Text-to-video will paint a prettier cone with a melted distal rim. Retouch the photograph. Then decide whether it ever needs to move.

GPT Image 1 Edit is 4K, 2 credits

GPT Image 1 Edit is OpenAI’s instruction-based image row: edit-image and image-to-image, image required, 4K ceiling, 2 credits, billed per call. Audio does not apply. No duration. The catalog job is the sentence you already have: describe the change in plain English and it repaints only what you asked, preserving the rest of the frame. Suited to precise product and layout touch-ups. That is not a second generate. If you re-prompt the hero from scratch, you do not have an edit. You have a new still that happens to mention a speculum.

The AI photo editor is the surface. Attach the photographed cone. Point at the object. “Kill the glare on the distal rim.” “Neutral grey card, keep the cone geometry.” “Remove the pallet label, do not redraw the taper.” Stop. A prompt that restages “a clinical close-up through an ear canal onto a tympanic membrane” is how the model “helps” by inventing anatomy. Anatomy is not a layout touch-up.

The row requires an image. Text alone is the wrong family. Max references listed: six. Use them as supporting views of the same speculum, not six unrelated ENT props. Aspects listed: 1:1, 2:3, 3:2, 16:9, 9:16. Qualities: SD, HD, 4K. None of those knobs mean “draw an eardrum.”

One change, then look. Stacking five instructions in one generate is how the taper averages into a cousin funnel. Two credits is cheap enough to be strict.

Overlay French size as authored type

French size is a claim. 2.5 mm. 4 mm. 5 mm. Whatever string the line card already prints. Video models do not typeset a medical-adjacent catalog number. Image models will try; they will also invent a character. Do not prompt “letter 4 mm on the flange” and ship the first misspelling. Do not ask the edit row to paint Fr. glyphs into the plastic.

If the photograph already carries a sharp, real size mark you printed, keep it. Zoom to 100% and read every glyph. Fine text is the first thing a generator loses. If the mark is missing or a lot scribble, burn the size after picture lock.

Add a text overlay burns one authored line for the whole clip — top, center, or bottom. It does not transcribe. “4 mm / Fr. 12” is overlay copy, not a prompt adjective. Overlay after the edit. Swap the string for the next SKU without regenerating the cone.

Do not:

• Prompt GPT Image 1 Edit to “also write the full size family across the card.”
• Ask Veo to put a French size on a keypad the speculum does not have.
• Generate a tympanic membrane, a cerumen scene, or an “after cleaning” canal as proof.

Wrong geometry on an ear speculum is a safety-adjacent miss. Wrong type is a listing miss. Both are cheaper to catch on a two-credit edit than on a motion pass.

/for/audiologists is the clinic, not this SKU

/for/audiologists is the clinic job: write to the daughter, not the denial. Fitting-room footage, a real switch-on with consent, captions because the audience cannot rely on the mix. That page is not an OEM catalog of disposable cones. Do not rewrite it. Do not steal its hook for a speculum orbit.

OEM content that books a distributor is the cone on a card and the size overlay. Clinic content that books a test is a named clinician, or an empty booth. Mixing them is how a speculum plate becomes a fake expert.

Synthetic presenters on health and finance are a YPP kill. A constructed audiologist prescribing a speculum size is a health-adjacent persona. Labelling the face AI does not leave the bucket. The OEM listing does not need a mouth. If someone must speak a protocol, that someone is a real named credentialed person on a different take. Generation can still do the speculum. It cannot be the exam, and it cannot be the tympanic membrane.

FAQ

Can GPT Image 1 Edit invent a tympanic membrane so buyers see “what the speculum looks through”?

No. That view is anatomy and a protocol. Photograph the empty cone. Overlay the French size. Do not invent a membrane so the SKU has a scene.

Why not text-to-image a “clinical otoscope set” if we lack a speculum photo?

Because the set is invented and the taper will drift. OEM is the cone you photographed. GPT Image 1 Edit requires that file: 4K, 2 credits, image in, one object named. Get the photo. Do not mint a patient to fill the gap.

Where does French size go if it cannot live in the prompt?

Type it. Add a text overlay after picture lock. Reports change; the cone should not. Prompted glyphs are a miss you regenerate to fix.

Is a talking AI audiologist acceptable if we disclose it?

Disclosure does not move a constructed health expert out of YouTube’s third bucket. Keep the speculum plate. If someone must speak, put a real named clinician on camera.