For teaching material the model ranking matters less than one rule: generate the illustration, never the fact. A generated diagram that is subtly wrong becomes something thirty students learn wrong, and unlike a marketing asset it does not get corrected by the market. Within that rule, AI video is genuinely useful for lessons and cheap enough that a term's worth of material costs a few dollars.
What it costs
Canon 8frame prices, a credit is $0.01 at pack rate. A six-minute lesson broken into four objective-sized videos:
| Element | Route | Credits |
|---|---|---|
| 12 illustrative beats | Veo 3.1 Lite, 8s silent at 33 | 396 |
| Narration, 900 words | ElevenLabs TTS, 13 per 1,000 characters | ~65 |
| Title and diagram stills | Seedream 5 at 6 | 60 |
| Music bed | Lyria 2 | 13 |
| Total | ~534 credits, about $5.34 |
The $19 Starter plan's 1,000 credits covers roughly two such lessons a month. Narration being effectively free is the part that matters most for teachers, because scripts get revised constantly and re-recording a human does not scale.
The accuracy rule, in detail
Generate: atmosphere, historical mood, abstract concepts, transitions, context shots, title cards, anything illustrative.
Never generate: diagrams with labels, maps, timelines with dates, chemical or mathematical notation, anatomical detail, or any image a student will copy into their notes as fact.
Generated text inside images is unreliable, which is exactly where a labelled diagram lives. The safe pattern is a generated background with the labels added in a real editor, or a diagram made properly and the generation used for the scene around it.
And say what was generated where a student might take an image for documentation. Historical footage in particular: a generated scene that looks like archive material in a history lesson teaches a false thing about evidence, independently of what it depicts.
Structure that produces retention
One learning objective per video, three to six minutes, demonstrate before explaining, and show the common mistake. Full structure in how to storyboard a training video, which applies directly to lessons.
The practical benefit of short single-objective videos in a classroom is not pedagogy, it is revision: students can find the two minutes they need instead of scrubbing a lecture.
School realities worth planning around
Policy. Many institutions now have AI-use policies covering teaching material. Check yours before building a term's content on it.
Accessibility. Captions are usually required, not optional. Generated narration plus a caption file handles this cheaply.
Device and bandwidth. Material watched on school wifi or student phones. Keep files modest and design for muted viewing.
Copyright of source material. Feeding a textbook image in as a reference is a rights question, not just a technical one.
FAQ
Is there a free AI video generator for teachers? Free tiers exist and mostly watermark or cap output. At around $5 a lesson, the paid route is usually simpler than working around a watermark, but check whether your institution has a licence already.
Can AI make a whole lesson? It can make the material. The sequencing, the accuracy check, and knowing where students get stuck are the teaching, and none of that is generated.
Should I use an AI presenter? For procedural and illustrative content it adds little. Your own voice over generated visuals is cheaper, faster to revise, and more trusted by students who know you.
Generate the illustration, never the fact. The 8frame canvas is free and unlimited, and generation is paid from $19/month.