Why video and audio are their own session
Video and voice are where AI feels most like magic - a brief becomes a talking avatar, a script becomes a narrated clip in minutes. That leverage is real. So is the exposure: a voice or a face is a person's identity, and synthetic media carries likeness, copyright, deepfake, and disclosure risk that text never did. This is the one session where the rights rail is loud on purpose. The move is the same as every session before it: the story and the edit stay human, and here so does the question of whose voice and face you are allowed to use.
Script and storyboard first, generate second 6 min live
Great AI video does not start in a video tool. It starts in a document: a written script and a storyboard. The generators - Runway-class video, ElevenLabs-class voice - are the production layer, and they are only as good as the story you hand them. Skip the writing and you get slick, generic clips nobody finishes. Write first, and the AI has something worth producing.
LiveThe script and the storyboard carry the piece3 min▶
A short video is a tiny story with a job to do. The script decides the words and the beats; the storyboard decides what the viewer sees in each of them. Both are cheap to write and expensive to fix after generation, so you do them first, on paper, before a single clip is rendered. AI can help you draft both - it is good at structure and options - but you pick the hook, the pacing, and the payoff.
- Script = the spine. Hook in the first two seconds, one clear value beat, one call to action. Short lines, spoken-word rhythm, no filler.
- Storyboard = the eyes. Shot by shot: what is on screen, the framing, the on-screen text, the transition. This is your shot list for the generator.
- Generate against a plan. With script and storyboard in hand, the video and voice tools produce to a brief instead of guessing - far less slop, far fewer regenerations.
The clip that was beautiful and unwatched. A team generated a run of polished AI videos straight from one-line prompts. They looked great and nobody finished them - no hook, no point, no reason to stay. When they wrote a real script and storyboard first and let the tools produce to it, completion rates jumped. The generator was never the problem; the missing story was.
Self-studyA short-video script skeleton you can reuse2 min read▶
Pin this shape for any 30-60 second short. It is the same hook-value-CTA spine you learned for social, tuned for the ear and the eye.
- 0-2s hook: the line or image that stops the scroll. Concrete, specific, a little tension.
- 2-40s value: one idea, delivered in short spoken lines, each paired with a shot and on-screen text.
- 40-45s CTA: one action - follow, read the full piece, try it. Not three.
- Under every line: a storyboard cell - shot, framing, on-screen text, transition - so production has no guesswork.
AI video + voice - with the rights rail loud 6 min live
Text-to-video, AI avatars, text-to-speech, and voice cloning are powerful and legally loaded. A voice and a face are a person's identity, protected by likeness and right-of-publicity law; footage and music carry copyright; and synthetic media of a real person is where deepfake harm and disclosure rules live. This is not a footnote here - it is the session. Never clone a real person's voice or face without clear permission, and disclose synthetic media where it matters.
LivePowerful tools, real exposure - the rules that keep you safe3 min▶
The generators do not check rights for you - that is your job, and in video and voice the stakes are higher than anywhere else in this course. The rule of thumb is simple: if a real person's identity is involved, you need their permission in writing, and if the media is synthetic, you disclose it.
- Voice and likeness are the person's. Cloning a real voice or generating a real face without documented, specific permission is a likeness and right-of-publicity problem - even if the tool makes it a one-click move.
- Inputs carry copyright. Footage, images, and music you feed in, or that a tool trained on, can be someone else's IP. Use rights-clear or licensed material and check the tool's commercial terms.
- Deepfake risk is reputational and legal. Realistic synthetic people are the highest-risk output you can make. When in doubt, do not.
- Disclose synthetic media. Where AI-generated or AI-altered people, voices, or endorsements appear, label them - platform rules and FTC-style disclosure are the norm, not the exception.
Self-studyThe rights guardrail, use by use2 min read▶
| Use | The rights guardrail |
|---|---|
| Text-to-video (generic scenes) | Check the tool's commercial terms; avoid recognizable real people, logos, and trademarks |
| AI avatars / presenters | Use licensed stock avatars or ones you have rights to; never a real person's face without written permission |
| Text-to-speech (synthetic voice) | Use the tool's licensed voices; confirm commercial use is allowed; disclose that the voice is synthetic |
| Voice cloning (a real voice) | Only with that person's documented, specific consent - never a public figure or a colleague on a whim |
| Music + stock footage | Licensed or rights-clear only; keep the licence on file |
| Any synthetic person or endorsement | Disclose it as AI-generated; follow platform and FTC-style rules |
When a row is uncertain, the answer is stop and clear it - not generate and hope. This table is your pre-publish checklist for every AI video and voice piece.
Repurpose an existing piece into video 4 min live
You do not start from a blank page. The Cadence long-form article you wrote in b4 is already a source of stories, and one idea can become text, social, and now video. Turning a written piece into a short-video script is the fastest, safest way onto the frontier - the thinking is done, you are just changing the medium.
LiveFrom one article to a short-video script2 min▶
Take the strongest single idea from the b4 article - one insight, one story, one number - and rebuild it for the ear and eye. A 1,200-word article is not a script; you extract its best 45 seconds and rewrite for spoken rhythm and on-screen visuals. Same idea, native to video. This is the multimodal-from-one-idea move you met in b3, now pointed at the highest-leverage format.
- Extract, do not narrate. Pull the one beat that carries on screen; do not read the article aloud.
- Rewrite for the ear. Short spoken lines, a hook up front, one CTA - not article prose.
- Storyboard the visuals. Each spoken line gets a shot and on-screen text; that is your production brief.
Script + storyboard a Cadence short ★ 13 min · everyone builds
Build the frontier piece from something you already have. Take an existing Cadence idea and turn it into a 30-60 second short-video script with a shot list and a storyboard - hook, value, CTA, ready to hand to a generator.
Pick one Cadence idea you have already used - from the brief in b2, the repurposing run in b3, or the b4 article. One idea, one clear point.
Ask a chat AI for a 45-second script with the three beats: a 2-second hook (give yourself two options), one value beat, one CTA. Spoken-word rhythm, short lines.
Ask it to add a storyboard cell under each line: shot, framing, on-screen text, transition. That is your shot list.
Now edit by hand: pick the hook, cut any filler, replace one generic line with a real Cadence specific, and make sure the CTA is a single action.
Read the script out loud against the storyboard. If it does not sound like a person and read cleanly shot by shot, fix it before any tool sees it.
Voiceover plan + a rights & disclosure check ★ 8 min · build your own
Before that short gets a voice, plan it and clear it. You will decide how the piece is narrated and run a synthetic-media rights and disclosure check you could defend to a client or a platform.
Decide the voice: your own recorded voice, a licensed synthetic voice from the tool, or - only with written consent - a cloned voice. Write down which and why.
Answer the three rights questions in writing: whose voice is this, whose face or likeness appears, and is each one cleared with documented permission.
Check the inputs: is the music licensed, is any footage rights-clear, does the tool's commercial-use term cover what you are doing.
Decide disclosure: if the voice or any person is synthetic, write the label you will show and where it goes. Where required, it is not optional.
Keep the completed check with the piece. If any answer is "not cleared", the piece does not publish until it is.
The voice that was not theirs to use. A team cloned a well-known creator's voice for a promo because the tool made it a two-minute job. It was fast, convincing, and a right-of-publicity problem the moment it published - a takedown and an apology followed. A thirty-second rights check would have caught it. On the frontier, the check is the work.
Try it yourself - this week ◐ 30-45 min total
- Finish both demos if you did not complete them live - a scripted, storyboarded Cadence short with a completed rights and disclosure check.
- Repurpose the b4 Cadence article into one 45-second short-video script. Extract the single best idea, do not narrate the whole piece.
- Fill in the one-page rights and disclosure check for that short. If any line is "not cleared", note what it would take to clear it.
- Write your default voice policy in one line: your own voice, licensed synthetic, or cloned-only-with-written-consent - and stick to it.
- Optional: if you have a video or voice tool, produce the first shot against your storyboard and feel how much the plan does the work.
Official sources covered
Taught from public AI content and video curricula, voice-tool practice, and the IP canon. Certificates and paid tooling stay official. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · Where does a good AI video start?
The script is the spine and the storyboard is the shot list. Generate against a plan and you get far less slop and far fewer regenerations. The story stays human.
2 · Before you clone a real person's voice or use their face, you must...
A voice and a face are a person's identity - likeness and right-of-publicity. Never clone a real person without written permission, and disclose synthetic media where required.
3 · The fastest safe way onto the video frontier is...
One idea becomes text, social, and video. The b4 article already did the thinking - extract the best 45 seconds and rewrite for the ear and eye. Same idea, native to video.