Air-classifier rotors keep vane count on MiniMax H3 (2K, 26cr)

Cyclone geometry already locked before swirl: this leftover is the classifier rotor, MiniMax H3 is 2K text-to-video, 5–15s, 26cr, audio null so vane count cannot drift, and cut size in µm is authored overlay rather than a generated d50.

A classifier listing that books is the rotor: vane count, cage diameter, the wheel a buyer will match to a spare. It is not the cyclone barrel you already photographed, and it is not a constructed mill tech walking d50. Write the rotor. Pick the 2K row. Leave the vessel to the cyclone post.

This leftover is the classifier rotor, not the cyclone

Cyclone geometry must lock before you add swirl already owns inlet, cone, and vortex finder: photograph the cut, then image-to-video swirl. That is a separator vessel. This leftover is the air-classifier wheel that sits upstream or in the same plant — vanes, a cage, a speed the cut sheet already named. Mixing them is how a rotor still becomes a fake cyclone.

Specifiers compare vane count. Twelve versus twenty-four is not styling. A generate that adds vanes mid-clip sold a different cut. Photograph the approved rotor when you have one. Text-to-video is the door when you do not have a still that must be frame one — a 21:9 hero, a wheel that does not exist as a photo yet.

Keep people off the generate. A constructed millwright explaining cut size is a presenter you did not hire. Prompt steel, light, and camera: locked wide, slow orbit of the cage, vanes readable, no new cyclone. Do not prompt “dramatic swirl through a cyclone then the classifier.” That paragraph restages two SKUs.

Manufacturers is plant-floor scale. This page is the classifier rotor. Do not rewrite the cyclone post as a wheel.

MiniMax H3 is 2K text-to-video, 5–15s, 26cr, audio null

Minimax H3 Text to Video is one catalog generate. MiniMax, text-to-video, 26 credits per second, max output 2K. Description: 2K video from a text prompt alone, durations from 5 to 15 seconds across six aspect ratios. requires_image is false. audio is null. Durations: 5s through 15s, every integer second. Aspects: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16. Qualities: 768P, 2K. That is the whole contract.

A 5-second take and a 15-second take are not the same bill. Floor is 130 credits. Ceiling is 390 credits. Pick the length first, then write one continuous beat that fits. Padding a 5-second rotor hold to 15 seconds is how the back half invents a second cage, extra vanes, or a cyclone you did not brief.

AI video generator is the door. Write the rotor. Select 2K when the plate will be paused as a still. 768P is for scouting. audio is null. H3 will not hiss mill air or invent a VO. Score it after. Do not prompt quoted micrometres and expect a title card. Start at 5s to see whether the vane count holds. Use 15s only after a short take already has the right cage.

H3 will invent a generic rotor. That is allowed when the ad is “we classify,” not “this serial-numbered wheel.” Your vane count on this photo is image-to-video, not this T2V slug.

Cut size in µm is overlay, not a generated d50

d50 is a datasheet claim. A model that typesets 45 µm on the cage is a rating you did not measure. Reports change. The rotor should not. Write the prompt as a clean plate: wheel in frame, no readable percent, no invented d50. Then put the real string on later.

Add a text overlay burns one authored line for the whole clip — top, center, or bottom. That is the cut-size line, the vane count, a CTA. It does not transcribe. Overlay after picture lock. Swap d50 45 µm for the next datasheet revision without regenerating the cage.

2K makes garbled in-frame glyphs more visible, not less. Keep type in the overlay. Do not ask H3 to letter micrometres onto a vane. Prompted d50 is a miss you regenerate to fix, and a generate that also rewrote the vane count sold a different machine while it was lettering.

The AI product video generator is the contract when a photographed rotor has to move: reference photos are the ground truth. Zoom to 100% on the vane count before you keep a take. Fine geometry is the first thing a generator loses. Prompt only the camera. Do not prompt a cyclone into existence so the ad “feels plant.”

Vane count cannot drift

The failure mode is a beautiful orbit of a different SKU. Twelve vanes becoming eighteen. Cage diameter walking. Those are claims. A drifted vane count sold a different d50 even if you never painted the number.

Prompt camera only. Do not write “make the rotor look more open.” Open is a different part. Do not first-last a 24-vane cage into a 12. Two different objects is a morph. H3 does not take a last frame anyway.

If you already photographed this rotor, you are not on this row. Animate the photo on image-to-video. H3 will redraw the vanes. Pause on the cage. Count the vanes. If they are the SKU, overlay d50 and ship. If they have been restaged, you bought weather.

FAQ

Does MiniMax H3 require a photo of our rotor?

No. The description is 2K video from a text prompt alone. If the photographed cage must be frame one, you are on an image-to-video row, not this one. Use T2V when a generic rotor is enough.

Will H3 add mill audio if I prompt “real classifier hiss”?

No. audio is null. Score later. Do not expect a native mix from a silent catalog flag. Cut size still goes on as overlay.

Can I text-to-video a cyclone and caption it as the classifier?

No. The cyclone post already owns inlet, cone, and finder. This leftover is the classifier rotor: vane count on MiniMax H3, 2K, 5–15s, 26 credits per second, audio null. A cyclone in the middle of a rotor shot is a cousin vessel.

Where does d50 go if it cannot live in the generate?

Type it. Overlay after picture lock. Cut size in µm is a datasheet string you typed, not a generated d50. Prompted micrometres are a miss you regenerate to fix, and they often arrive with a drifted vane count.