Print farms: time-lapse is a speed edit, not a generate

If the camera already ran for four hours over the bed, you do not have a generation problem. You have a playback-speed problem. A time-lapse of a real print is the same file, faster. Change a video's playback speed is the job: describe the factor, preview at 480p, export. Do not spend a video model inventing a printer you already own.

The AI video generator is for a shot you do not have. A fake gantry, a fake nozzle, a fake SKU rising out of a plate you never printed is a different object. People who buy prints can see layer lines. A generated farm that "looks like" your enclosure is a product video of a machine that is not yours.

The printer already filmed itself

Lock a phone or a webcam. Same angle, same light, whole job. That is the source. Failures, stringing, a pause at layer 400 — those are also real, and they are why operators used to throw the file away. You do not need the failed four hours in the cut. You need the keeper minutes, sped.

edit_video treats speed as a described EDL instruction, not a keyframe graph. "Play this clip at 8×" or "4x time-lapse on the whole file." Fine-grained curves are not the interface; preview is how you check. The 480p preview is free with a per-user cooldown. The final is the editor export.

You can speed a range and leave the rest at 1× — a long build, then a real-time lift off the plate. Describe the range and the factor. The rest of the clip can stay normal.

Audio stretches or compresses with the picture unless you say otherwise. A printer time-lapse does not need intelligible speech. If you later add VO ("six hours, 0.2 mm, PETG"), add it after the ramp, against the new clock. Silence on pitch is how a stray room mic survives a speed-up as a cartoon, on a file you otherwise liked.

Speed is an edit, not a second printer

Generation interpolates toward a guess of how a printer moves. Real toolpaths are uglier and more convincing. If you already have the footage, changing playback speed keeps:

Your actual enclosure, not a stock one.
Your actual part, not a dragon the model likes.
Your actual lighting, which is the farm's lighting.

A generated time-lapse is a first-last-frame or text-to-video job pretending to be documentation. Use that only when there is no plate. Even then, do not put a fake printer in a "this is our shop" reel. Portfolio and visualizer are different libraries.

If the source is three minutes and you need ten seconds, that is a factor, not a model. If the source is a still of the finished part, that is image-to-video of the object, not a fake farm. Do not caption real-time then speed: timed text assumes the duration after the ramp.

Approve the take at 1× first

Order:

Approve the recording at real time. If the product on the bed is the wrong SKU, 8× will not help.
Preview the ramp on that file. Check pitch if any audio stays.
Export. Then music and captions against the new duration.

A true time-lapse of a long event is a parent you already have. It is not an 8× panic on a 6-second generate you hoped would look like a farm. The generator is still useful around the farm — a finished-part turn, a commission still you have not printed. None of those require a synthetic gantry wearing your logo.

FAQ

I have a four-hour recording. Do I generate a 6-second loop instead?

No. Change playback speed on the recording. Generation is for a plate you do not have.

Will speeding up make a bad print look better?

It will make the same print occupy fewer seconds. Warping and spaghetti stay in the frames. Trim the miss, or reshoot. Do not ask a model to invent a clean farm over it.

Can I generate the printer if the camera moved?

Fix the source next time — lock the camera. A generate of a "perfect" overhead is a different SKU. If you only have a still of the finished part, animate that still. Do not fake the machine.

Does preview apply the speed change?

Yes, at 480p. Use it to inspect the keeper's ramp. Then pay the editor export once.