Voice Cloning, Consent and Disclosure

The lesson that is not about craft

Everything so far has been about making better videos. This one is about not causing harm with them, and about not ending up in a situation you cannot talk your way out of.

The tools in this course can copy a voice from a short recording. That capability is already being used to call parents with a child's voice asking for money, and to authorise payments in the voice of a company director. When you use a voice tool, you are handling something that belongs to a person. Treat it accordingly.

Cloning a voice requires that person's consent

The rule is short. Never clone a real person's voice without their written permission.

Not your colleague who probably would not mind. Not your friend, as a joke. Not a public figure, because they are famous. Not a family member who has died, however loving the intention, unless the people entitled to decide have agreed. Fame is not permission. Availability of recordings is not permission. Only the person is.

Written matters because memory is unreliable and relationships change. A short message that says what you may do, in what, and for how long, protects both of you. Keep it. If you would be uncomfortable showing that message to the person later, you did not have consent.

Consent is specific, limited and revocable

Permission to narrate one training video is not permission to build a permanent voice for your marketing. Ask for what you actually need, and name it: which project, which platforms, how long, and what the voice may never be used for.

Say plainly what happens if they change their mind. A person who agreed last year may want out this year, and the reasonable answer is that you stop using it and delete the model. Agreeing that up front is far easier than negotiating it during a disagreement.

The same applies to faces. If a real person appears in anything generated or altered, their likeness is theirs, and the same standard applies.

Disclosure: when to say it is AI

Platform rules are moving quickly, and they differ. The general direction is consistent across the major ones: realistic synthetic content that could be mistaken for a real recording is expected to be labelled, most platforms provide a toggle or field for it when you upload, and some apply automatic labels when they detect generation. Rules are usually stricter for anything touching news, elections, public figures, health, or money.

Because those rules change, this lesson will not list them. Find the current policy for each platform you publish on, and check it again occasionally.

A workable rule of thumb while you are deciding: if a reasonable viewer could believe a real person was filmed or recorded, label it. A synthetic narrator over generated footage should be labelled. A generated shot of rain on a window under your own voice sits at the mild end, and saying so anyway costs you nothing.

Put the label where it can be seen: use the platform's own setting, and add a plain line in the description such as "Voiceover and footage generated with AI." Audiences are far more forgiving of disclosed AI than of AI they discover later.

Music, footage and other people's rights

Popular music is not yours to use because you bought it or because the clip is short. Use the platform's licensed audio library, or music explicitly licensed for the use you have in mind.

Check the licence on the tools themselves. Free tiers often permit personal use only, and quietly exclude commercial use, which is exactly the use a small business needs. Read what your plan allows for the output, not just what it allows you to make.

Stock footage carries conditions too. Restrictions on sensitive topics and on implying that a person shown endorses your product are common.

I would like to use a synthetic copy of your voice.
What for: [project name and what it is].
Where it will appear: [platforms or places].
How long: until [date], after which I stop using it.
It will never be used for: [advertising, political content,
anything financial, anything you have not seen].
You can withdraw permission at any time by telling me,
and I will stop using it and delete the voice model.
Please reply with I agree, and today's date, if you are happy with this.

Write three one-line AI disclosure statements for a video description.
Context: [what is generated: voice, footage, both].
Audience: [who they are]. Tone: plain and factual, not legalistic.
Each must be under 15 words and understandable by someone
who knows nothing about AI. No jargon, no marketing language.

Act as a cautious reviewer before I publish this video.
List every element that needs permission or a licence: voices,
recognisable people, music, stock footage, logos, product names.
For each one, ask me the single question I need to answer
to confirm I am allowed to use it. Do not reassure me.
[describe your video and every asset in it]