A 60-second explainer should leave the viewer thinking “oh, that’s what that means” — not “what was that?” The failure pattern with AI explainers is always the same: people start with a cool visual, write the script to fit, and the concept never lands. This tutorial flips the order. Script first, then storyboard, then generate visuals that serve the script, then voice last. Sora 2 and Veo 3.1 do the heavy visual lifting; ElevenLabs or a built-in TTS handles narration. The result is a 60-second piece a viewer can watch once and then explain to someone else.
TL;DR
Write one concept as a 3-beat script (hook 10s / body 40s / payoff 10s, roughly 150 spoken words). Storyboard 6-8 shots, one per phrase of narration. Both Sora 2 and Veo 3.1 generate clips in 8-10 second chunks at 1080p, so a 60-second piece is six to eight stitched generations, not one render. Keep one visual register the whole way through. Narrate with an ElevenLabs voice clone (Starter $5/mo and up, as of June 2026) or a hired actor for brand work. Cut on phrase boundaries, keep music at -18 dB, and put the payoff sentence on screen as text. Budget about two hours end to end.
What this covers
A script-first explainer workflow: one clear concept, a 3-beat script (hook, body, payoff), a 6-8 shot storyboard, generated visuals that match each beat, and a clean narration mix. Tools: Sora 2 (inside ChatGPT) or Veo 3.1 (inside the Gemini app or Google Flow) for visuals, ElevenLabs or your preferred TTS for narration, and any editor for assembly.
Who this is for
Educators turning lessons into shareable shorts, founders explaining their product to a cold audience, content creators with hard-to-grasp ideas to teach, and consultants who need to explain a concept once and reuse it across decks.
When to reach for it
Product education videos, onboarding intros, social-media explainer posts, course-trailer videos, internal training shorts, and any pitch where the audience does not have your jargon yet.
Pick your visual tool (June 2026)
Both leading models top out at short clips, so a 60-second video is always several generations stitched together. Plan your storyboard around that limit rather than fighting it.
| Tool | Where you use it | Clip length | Max resolution | Plan to start |
|---|---|---|---|---|
| Sora 2 | ChatGPT (Plus $20/mo, Pro $200/mo) | ~10s, up to 20s on Pro | 1080p | ChatGPT Plus $20/mo; free tier can no longer generate video |
| Veo 3.1 | Gemini app + Google Flow | 8s, chain clips for longer | 720p / 1080p (4K via Vertex/API) | Google AI Pro $19.99/mo; AI Ultra $99.99/mo for higher limits |
Notes as of June 2026: free ChatGPT users can no longer generate Sora video. Veo 3.1 in the Gemini app supports native 9:16 and a scene-extension feature that chains clips, which is handy for a continuous explainer. If you only generate through the API, Sora 2 runs about $0.10/sec at 720p (sora-2-pro $0.30-0.50/sec); Veo 3.1 ranges roughly $0.03-0.40/sec depending on tier and audio.
Before you start
- Write the one-sentence concept. If you cannot fit it in one sentence, the video is too broad — split it.
- Identify the audience’s prior knowledge. A 60-second video assumes a level of context; name what you assume.
- Pick the metaphor that explains the concept. AI visuals work best when the underlying metaphor is concrete: a leaky bucket, two stacked boxes, a forked road.
- Decide narration language and tone before generating any visuals. Visuals follow voice, not the other way around.
Step by step
- Write a 3-beat script: hook (10s — name the problem or surprise), body (40s — explain via metaphor), payoff (10s — one sentence the viewer can repeat). Time it aloud; 60 seconds is roughly 150 spoken words.
- Storyboard 6-8 shots. Each shot serves one phrase of narration, not one sentence. Match phrase rhythm to cut rhythm. Because each generation is only 8-10 seconds, your storyboard count and your generation count line up almost one to one.
- For each shot, write a prompt that depicts the metaphor literally. Avoid “an illustration of X” — describe the action:
a single water drop hits a glass surface and ripples outward, top-down view, soft daylight, slow motion. - Generate at a consistent style. Pick one visual register (clean 3D, paper cutout, photographic) and stay there. Mixed registers in a 60-second piece look like the model broke. If you switch between Sora 2 and Veo 3.1 mid-project, expect a register jump and re-prompt to match.
- Record narration. ElevenLabs instant voice clones land naturally with about a minute of clean source audio (Starter plan, $5/mo, June 2026); Professional Voice Cloning on Creator ($22/mo) holds up better across longer passages. For a true human feel, hire a voice actor.
- Assemble: narration on top, visuals cut to phrase boundaries, light music bed at -18 dB. Music should be present but not commenting on the script.
First-run exercise
- Pick one concept you can explain in person in two minutes. Write a 150-word version that fits 60 seconds when read aloud.
- Storyboard it in pen — 6 panels, one per script beat. Do not skip this; sketching on paper is faster than re-prompting.
- Generate one shot in three different style registers. Pick the one that best serves the script, then regenerate the other shots in that register.
- Read the narration into your phone before going to TTS. Hearing your own pacing exposes script weaknesses faster than any AI critique.
Quality check
- A viewer who did not know the concept can repeat it back after one watch. Test on someone.
- Visual register is consistent across all shots. No “this one looks 3D and that one looks 2D.”
- Cuts land on phrase boundaries, not mid-clause. Re-cut anything that breaks reading rhythm.
- Music bed sits below narration. If you can hum the music line, it is too loud.
- The payoff sentence is on screen as text, not just spoken. Sticky concepts get re-read.
How to reuse this workflow
- Save the 3-beat script template as a Notion or doc snippet. A new concept slots into hook / body / payoff in 10 minutes.
- Build a metaphor library for your domain. Reusable metaphors (leaky bucket, fork in the road, stack of boxes) explain dozens of concepts.
- Keep a style-register preset of three or four prompts that always come out in your house style. The next explainer reuses the same register.
- Maintain one voice clone or one hired actor across the series. Voice consistency is the cheapest production-value lift you have.
Common mistakes
- Picking the concept after seeing a cool AI visual. Concept first, visuals second.
- Two concepts in one video. Split them; 60 seconds carries one idea well, not two.
- Inconsistent visual register. Mixing realistic and stylized in the same video reads as broken.
- Treating a 60-second video as one render. Both Sora 2 and Veo 3.1 cap at short clips, so storyboard for stitched generations from the start.
- Running default TTS straight through without tuning. Slow it down, add pauses, re-record troublesome lines.
- Music loud enough to compete with narration. Narration is the script; music is wallpaper.
- No on-screen payoff text. Viewers re-read; if the payoff is only audio, it does not stick.
FAQ
- Sora 2 vs Veo 3.1 for explainers?: Sora 2 handles abstract metaphor visuals well and is built into ChatGPT Plus ($20/mo). Veo 3.1 (Gemini app, Google AI Pro $19.99/mo) handles human-presence shots and longer continuous scenes better thanks to clip chaining, and it can push to 4K via the API. Pick one per project; mixing them mid-video usually causes a visual register jump.
- How long can one AI clip be?: As of June 2026, Veo 3.1 generates 8-second clips (chain them for longer), and Sora 2 generates roughly 10 seconds, up to 20 on the $200/mo Pro tier. A 60-second explainer is six to eight stitched generations.
- Should I write the script or have AI write it?: AI for the first draft, you for the rewrite. Final scripts almost always need a human pass for voice and pacing.
- How long does a 60-second explainer take?: Script 20 minutes, storyboard 10, visuals 30-60 (across multiple short generations), narration and edit 30. Budget about two hours end to end.
- Is AI narration good enough?: For utility content, yes — an ElevenLabs voice clone (Starter $5/mo) reads clean copy well. For brand and emotional content, hire a voice actor; the difference is audible in five seconds.
- Best aspect ratio?: Make 9:16 for social and 16:9 for embed. Veo 3.1 generates native 9:16, so you avoid awkward cropping. Same script, two exports, no extra script work.
Related
- AI Short Form Video Prompts
- AI Video Storyboard Workflow
- AI Short Video Script Tutorial
- AI Video Prompt Basics
- OpenAI Sora help center
Tags: #sora #veo #explainer #Tutorial