Correction, 2026-10-01: the first version of this page said a five-minute ElevenLabs narration costs "a few cents". ElevenLabs TTS on 8frame is 13 credits per started 1,000 characters; a five-minute script is about 750 words, roughly 4,000 to 5,000 characters, so it is 52 to 65 credits, about $0.52 to $0.65. Still small, but not cents.
One learning objective per video. That single rule fixes most training video problems, because the failure mode of the format is a twenty-minute film covering nine things where the viewer retains two. Board each objective as its own short piece, and a course becomes a set of findable, rewatchable units instead of a monolith nobody scrubs through.
The structure per objective
1. Why this matters, in one sentence. Not a preamble. The consequence of not knowing it.
2. Demonstrate first. Show the whole task at normal speed before explaining any step. Learners need the shape before the detail, and explaining first produces viewers who follow along without understanding what they are building toward.
3. Break it into steps. Three to six, each with a visible state change so the learner knows a step completed.
4. Show the common mistake. The most valuable thirty seconds in most training videos and the most commonly omitted.
5. Recap the shape, not the steps. A summary that repeats all six steps is a second video.
Three to six minutes per objective. Beyond that, split.
Chunking rules that hold
One idea per screen. If the frame shows two things, the viewer picks one.
Pause after each state change. A beat of nothing is where comprehension happens; continuous narration prevents it.
Repeat the objective at the end. Not the content, the objective.
Make it scrubbable. Chapter markers, on-screen step numbers, and a title card per step, so someone returning to find step four can find step four.
What to generate
Training content has a specific risk: it is treated as authoritative, and errors in it propagate into how people do their jobs.
Generate: context and environment shots, abstract illustrations of concepts, transitions, and title backgrounds. Narration is a good fit here too, since training scripts are long and revised often: ElevenLabs TTS is 13 credits per started 1,000 characters. At about 150 words a minute, a five-minute narration is roughly 750 words, 4,000 to 5,000 characters, so 52 to 65 credits (about $0.52 to $0.65, and under the 5,000-character limit of one generation). A script change is a re-render rather than a re-record.
Do not generate: the procedure itself, the interface, safety-critical demonstrations, or anything a learner will copy exactly. Record real screens and real hands. A generated depiction of a procedure that is subtly wrong is worse than no video.
Say what is generated where a learner might otherwise take a generated illustration for documentation.
FAQ
How long should a training video be? Three to six minutes per objective. Longer material becomes a series, which is also easier to update.
Should training videos have a presenter on camera? For soft skills and context, yes, it helps. For procedural content, the screen or the hands are the subject and a talking head competes with them.
Can AI narrate training content? Yes, and it is one of the better fits, because scripts change and re-recording a human is expensive. Have a person review every language before it ships.
Sources
- Speaking rate: National Center for Voice and Speech, Voice Qualities ("about 150 words per minute"), checked 2026-10-01
- 8frame credit prices (ElevenLabs TTS, 13 per started 1,000 characters, up to 5,000 characters per generation) and plans: the 8frame tool registry and content fact sheet, checked 2026-10-01
One objective, demonstrated before explained, with the common mistake shown.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.