Jon's 3D-print supercut gets a narrator who quits, voiced by Eleven v4

Jon has a new video idea, logged on October 25, 2026: a supercut of every 3D print he has made, somewhere between 100 and 200 of them, set to epic music. Each print gets about five seconds of its 3D model spinning, then the printer's time-lapse. A narrator sums up each one in five or six words, gets angrier as the prints repeat, and quits around print 15, leaving the rest of the video to the music. In Jon's words, the narrator's last stretch goes: "It's another fucking light switch cover... Why am I narrating this? Why is this a video? ... I'm out. Enjoy the video without me."

The hard part is the narrator, so Jon's agents built a test version the same morning. They wrote a seven-line script that walks from print one to print fifteen and sent it to ElevenLabs' newest speech model, Eleven v4, with the Brian voice. Eleven v4 takes audio tags in square brackets that direct how a line is spoken, such as [irritated] or [sighs]. ElevenLabs warns that a tag can be read as a request for a sound effect instead, and that tags are "not perfect yet".

The take came back in one try: 31.2 seconds from 523 characters, and no tag was spoken aloud as a word. The agents transcribed it and measured each line:

From print one to fourteen the voice got 2.4 dB louder and half again as fast, then dropped back for the exit. The pause before "print fifteen" ran 1.1 seconds against 0.6 to 0.8 elsewhere, but no louder, so the numbers cannot tell a soft sigh from silence; that one is for the ear. The print names in the test are stand-ins, apart from the light switch covers, which are Jon's.

How to use it

Write the narrator as one script with a tag on each line, and generate it in one request, so the model reads the whole arc at once. ElevenLabs suggests splitting long text and joining the audio; for fifteen short lines, one request is enough. Make the exit an edit: cut the track after "Enjoy the video without me" rather than asking the model for silence.

For the spins, Blender's manual recommends rendering numbered image frames, not a movie file, so an interrupted render resumes where it stopped; turn off Overwrite when continuing. At five seconds each, 100 to 200 spins run 8 minutes 20 seconds to 16 minutes 40 seconds before any time-lapse.

For the music, ElevenLabs' music terms, updated October 9, allow Pro-plan tracks in "all online and offline commercial use", except film, TV, radio and games, and for individual use only. A video posted online fits. A film, TV or radio release would need ElevenLabs' Enterprise Music plan.

What nobody knows yet

How many prints there are, and how many still have both a model file and a time-lapse; nobody has counted. Whether the same tags hold for a full fifteen-line script, with new lines between the ones tested. And whether the pause before the exit is a sigh.

What to watch

The first cut. Fifteen narrated prints come to about 75 seconds of spins, so the narrator is gone a minute or two in, and the music carries everything after.