Video & voice

Why add video and voice

Photos show what your persona looks like. Video and voice show who she is: how she moves, laughs and talks. You don't need either on day one, but they make a persona feel alive.

There are three kinds of tools:

1. Video generators turn a text prompt or an image into a short clip.
2. Talking-head tools make a face speak in sync with an audio track.
3. Voice tools turn text into speech (often called TTS), or copy a voice you have the right to use (cloning).

Tools change fast. The examples below were checked on 28 September 2026. Features, prices and terms change often, so re-read the terms before you rely on any tool. As with images, pick tools whose terms allow what you plan to do, because your persona's job includes promoting casino and sportsbook campaigns.

Video generators

Heads-up: OpenAI's Sora is gone. The app closed on 26 April 2026 and the API on 24 September 2026, so older tutorials that use it no longer apply (OpenAI Help (opens in a new tab)).

Talking-head tools

• Hedra makes a talking character from an image plus audio, and has a free plan (pricing (opens in a new tab)). No gambling clause found; it bans explicit content and deepfakes without consent (policy (opens in a new tab)).
• D-ID turns a photo into a talking avatar. Its terms restrict only illegal gambling, and every animation carries a synthetic mark (terms (opens in a new tab)).

Flagged: these terms prohibit gambling promotion.

• HeyGen bans promoting goods and services related to online gambling (HeyGen (opens in a new tab)).
• Synthesia bans promoting or advertising age-restricted goods, including gambling (Synthesia (opens in a new tab)).
• Captions (Mirage) effectively bans gambling content, and also bans "suggestive" inputs (Mirage (opens in a new tab)).

Voice tools

• Cartesia: text-to-speech with instant voice cloning, and a free tier (pricing (opens in a new tab)). Other people's voices only with their explicit consent (policy (opens in a new tab)).
• Murf: text-to-speech and voiceover. Cloning requires the speaker's explicit written consent (terms (opens in a new tab)).
• Flagged: ElevenLabs bans using it to facilitate real-money gambling (ElevenLabs (opens in a new tab)).

Give her an original voice

Write a short voice description to match your persona card. For Alexis: warm, playful, a little husky, never breathy. Then pick a stock voice in your tool that matches it.

If you want a unique cloned voice, only clone a voice you have permission to use, such as a voice actor who has signed a consent form. Never clone a celebrity, a friend or anyone else without it. Cartesia requires explicit consent and Murf requires it in writing (Cartesia (opens in a new tab), Murf (opens in a new tab)).

Make your first test clip

1. Start from a reference image from your casting set. Image-to-video keeps the face closer to your reference than a text prompt alone.
2. Keep the motion simple. For example: Alexis turns to the camera and laughs, city lights behind her, handheld phone style. Start small; you can get more ambitious once you know how the tool behaves.
3. Generate a few versions and keep the one that looks most like her.
4. Add a line of speech. Write about 15 words in her voice, generate the audio, then lip-sync it with a talking-head tool.
5. Label it. Add an AI disclosure label: a caption or on-screen note that says the clip is AI-generated.

Save your settings next to your image recipe so the next clip matches.

Labels for video and audio

Realistic AI video and voice are exactly what platforms want labelled:

• Meta requires people to disclose photorealistic video or realistic-sounding audio that was digitally created, applies an "AI info" label, and may apply penalties if you don't (Meta, Feb 2024 (opens in a new tab), Meta, Apr 2024 (opens in a new tab)).
• TikTok requires a label on AI content that shows realistic-looking people or scenes. You can use its AI-generated content label, or a clear caption, sticker or watermark (TikTok (opens in a new tab)).
• EU: from 2 August 2026, the EU AI Act requires people who create deepfakes to disclose that the content is AI-generated. Whether a fully fictional persona counts isn't settled, so label anyway (European Commission (opens in a new tab)).

Promo videos need extra care

If a video will promote a campaign, make it with tools whose terms allow that. Then disclose it properly: the FTC says that in video you should both say and show the disclosure, and repeat it from time to time in live streams (FTC (opens in a new tab)). Lesson 7 covers where promo videos can and can't go.

Next lesson: setting up her accounts the right way.