You need someone on screen saying words, and you don't have a set, a crew, or that person's afternoon. There are four routes: generate a face, clone yourself, cast a licensed actor, or film it. They differ on cost and quality — and, the part most guides skip, on whether you're allowed to publish the result.
What are the four approaches?
Each solves a different problem, and the differences matter more than they first appear.
Generate a face. Prompt a face, then drive it with your script. Immediate, cheap, and the output can be convincing.
Clone yourself. Build a persistent AI twin from your own photos, then cast it in everything afterwards.
Cast a licensed actor. Pick a real person from a library who signed a licence for exactly this use.
Film it. Still the best option when you have the budget and the schedule.
| Approach | Setup | Consent Trail | Best For |
|---|---|---|---|
| Generate a face | Immediate | None | Background, non-commercial |
| Clone yourself | One-time | Yours by definition | Founder video, presenting |
| Licensed actor | Immediate | Signed | Client work, commercial |
| Film it | Highest | Yes | Anything with budget |
Consent trail is the column most comparisons omit, and the one that decides whether a brand will sign off on your work.
When should I generate a face?
For background characters, crowd scenes, personal projects, and rapid testing. It's fast and it's cheap.
Where it falls short: you don't know who that face resembles. If it happens to look like a real person and you use it commercially, the fact that a model generated it doesn't resolve the likeness question. Intent isn't the test — resemblance and commercial use are.
Fine for a personal project. Risky for client work.
When should I clone myself?
If the person speaking is you or your founder, cloning is the obvious answer. You consent by definition, the setup happens once, and every video afterwards casts the same recognisable face.
It typically needs twenty or more photos across varied angles, and a few minutes of training. After that it's reusable indefinitely.
Where it falls short: you still look like you. If the story needs a different age, build or type, cloning doesn't help.
When do I need a licensed actor?
For anything commercial, and anything a client or brand approves.
A licensed actor is a real person who signed a licence for this specific use. You know whose face it is and you know what you're permitted to do with it. On HexCoded, those actors are paid every time they're cast, with consent signed and liveness-verified.
Where it falls short: you're choosing from a library rather than casting exactly the face in your head. And it costs more, because a person is being paid.
One thing to check in any likeness licence: whether it covers paid advertising. It's the most commonly excluded use and the most commonly assumed.
What makes the performance actually work?
Whichever route you pick, the performance is where these fail. Four things matter.
Keep lines under twelve words. Longer lines drift out of sync. It's a short drama convention for exactly this reason.
Write the pause. Models fill the space after a line with nothing unless told. Specify what happens when the line ends — a held look, a blink, a turn away.
Match the language to the actor. A natively delivered line reads differently from a dubbed one, especially in close-up.
Don't over-direct the emotion. Describe behaviour — "jaw tight, doesn't raise her voice" — rather than naming the feeling.
- Personal project or background character → generate a face
- You or your founder speaking → clone once, reuse forever
- Client work, commercial, brand-approved → licensed actor
- Budget and schedule available → film it
- Whatever you pick: lines under twelve words, and write the pause
Cast an actor and type your script
Real licensed actors, AI actors, or your own twin.
Open Talking ActorsOnly with their consent. That's what a licensed actor library provides — people who agreed to exactly this use. Doing it without consent is a legal problem regardless of the tool.
It has no consent trail, and if it resembles a real person the likeness question doesn't disappear because a model produced it.
Generated is cheapest per clip. Cloning is cheapest across many videos, since the setup cost is one-time.
Usually long lines. Keep dialogue under twelve words — longer lines give the model more time to lose alignment.
It depends on your licence's media clause and the platform's own AI rules. Check both — paid advertising is the most commonly excluded use in likeness licences.