Cryostats: chamber still, never generated customer tissue

Photograph the empty chamber. I2V after identity holds. Generating a customer's specimen is a likeness-and-lab problem.

A cryostat listing that ships is the head you sell: knife, chuck, anti-roll plate, window, the empty well a histotech already knows. It is not a slice of someone else's tissue. Photograph the empty chamber. Animate that photograph. Leave customer specimens off the prompt.

Photograph the empty chamber

The buyer is matching a head they will install. They need this cryostat: the cryobar, the window gasket, the handwheel, the serial plate if the serial plate is the claim. A paragraph that says "medical cryostat cutting frozen section, cinematic lab" is a request for a cousin cabinet and a cousin specimen. The cousin is a failed SKU. The specimen is worse.

Shoot the unit empty. Door open or window clear enough that the well reads. If a service photo still has a chucked block in it, reshoot or crop until the block is gone. Motion will not un-cut a section you left in frame.

AI image to video is the aisle once that jpeg would pass a listing audit. Frame one is the file. The prompt is camera and physics — a slow push on the knife, a hold on the empty well, frost already on the glass if frost was already there. Re-describing "a busy pathology lab at golden hour" is how the model starts drawing a second head and a second case.

Do not prompt a ribbon, a cassette, a stained slide, or a "typical breast biopsy." Those are claims about tissue you do not own. Empty is honest. Occupied is a lab problem wearing a product brief.

Image-to-video after the head is yours

Seedance 2.5 is the still-to-motion family: four to thirty seconds, 720p on the I2V job, native audio, forty-seven credits listed. Image-to-video is the still-to-motion job on that family; the published URL is this slug. Thirty seconds in one pass is why a SKU holds. Stitching four short text-to-video clips around a knife is how the chuck drifts.

Prompt only the delta. Locked camera. No new window. No new person. Write the mix if you need one — compressor hum, a quiet room — or write that you want no speech and no music. Native audio is on. A silent brief still gets a bed you did not choose.

Pick 5s or 8s when the still only needs a hold. Do not pick 30s to "find" a prettier chamber. 720p is the ceiling on that endpoint.

Kling 2.5 Turbo is a different row: text-to-video, 5s or 10s, 1080p, 21 credits, audio unset, no image required. Unset means do not claim native audio. Fast quality balance. That is generic ice B-roll you can throw away when the sentence is still wet. It is not the cryostat you stock. Turbo will invent a cousin mill and, if you mention sections, a specimen. Leave Turbo. Open image-to-video with the photo.

Customer tissue is a likeness-and-lab problem

A generated section is not "medical atmosphere." It is a picture of tissue. If that picture is a customer's specimen, or a lookalike of one, you have a likeness problem and a lab problem in the same frame.

A deepfake of a private person is not a style. A patient is private. A block with a case number, a cassette that names a clinic, a face in the window during a live cut — those are people and records. TikTok's AI-generated content help forbids adult private figures used without permission. Minors are a hard stop. Labelling the clip AI-generated does not create a grant. "We changed the stain" does not either.

You do not get a carve-out because the subject is a slice instead of a portrait. A specimen is still that person's tissue. A synthetic "typical" biopsy seeded from a real case photo is a replica of a private figure's medical material. Style reference of a case still is how you get their case. Do not.

What is allowed: the empty head you photographed; frost; knife motion; a chuck with no block. What is not: a photoreal histotech, a generated patient, a ribbon that "looks like" last Tuesday's case.

Synthetic presenters on health and finance are a YPP kill. Overlay a model string after picture lock if the listing needs it. Do not mint a twin of the tech.

What this is not

It is not text-to-video "frozen section, cinematic, 4K." That invents the head and the tissue together. Kling 2.5 Turbo will do it in 5s for 21 credits. The file will be wrong.

It is not a one-still wiggle that leaves a cassette sitting in the well "for scale." If the well is empty, keep it empty. If a real demo still has a training block you own, name it as a training block or leave it out.

It is not a talking host. Type the setpoint and catalog number as overlay after the take if those strings must stay true. Serial plates that matter stay in the photograph.

Inspect the midpoint. If the middle grows a pink ribbon or a face in the glass, reroll the motion. There should be no specimen.

Would a histotech who already runs this head say the clip is this empty chamber? If not, you generated a lab. Photograph the well. Animate the well.

FAQ

Can I prompt Kling 2.5 Turbo for "our cryostat" and skip the photo?

No. Kling 2.5 Turbo is text-to-video, 5s or 10s, 1080p, 21 credits, no still required. It will invent a cousin cabinet. Photograph the empty chamber. Run image-to-video on Seedance 2.5 after that plate holds.

If we label the video AI-generated, can we add a photoreal section "for realism"?

Not if that section is a customer's specimen, a lookalike of one, or a still you pulled from a case file. Labelling tells viewers the media is synthetic. It does not create a grant, and it does not make a lab record into atmosphere.

Does Seedance 2.5 I2V use a separate folded model URL?

No. The published family page is Seedance 2.5. I2V is the still-to-motion job on that family: 4–30s, 720p, 47 credits listed, native audio. Launch it from AI image to video.

Do we need a talking histotech to explain the chamber?

No. Empty-head motion plus authored type is the listing. A synthetic clinician is a health-persona problem you did not need in order to show a knife and a well.