How to record your digital twin: script, light and sound in seven minutes

You need a table, a camera on a tripod, soft light from the front, the microphone on and about seven minutes. You do not need a professional studio or to memorise anything: the idea is to record yourself as if you were in a meeting.

The twin learns from a video of you. Its face, its gestures and its voice all come from that same recording, so the quality of the video matters more than its length. One good seven-minute take is worth more than an hour of badly lit video.

The script has four parts back to back, and each one teaches the twin something different. This table sums up what you do in each and how long it lasts.

That comes to about 6 minutes 40 seconds. You can record it all as one continuous video and then mark where each part starts, or record three separate videos: one listening, one speaking and one with the silent gestures.

Like a well-run video call. These are the framing rules, one by one.

• Landscape video (16:9). Never portrait.
• Seated at a table, from the waist up, with your hands resting on the table.
• Camera at the height of your eyes, 1.5 to 2 metres away. Look at the lens, not at the screen.
• A still background: a bookshelf, a wall or a plant. No people or screens switched on behind you.
• Your hands never pass in front of your mouth or chin.

The last rule is the one most often forgotten. When a hand covers your mouth, the twin loses the reference for how your lips move as you speak.

Soft light from the front and a fixed camera. A window works, as long as there is no direct sun.

• Tripod: nothing hand-held.
• Soft light from the front, no direct sun.
• 1080p or 4K, at 25 or 30 frames per second.
• Lock exposure and focus before you start, so the image does not change halfway through.
• Microphone on: the twin needs your voice.

A recent phone on a tripod covers all of this. What does not work well is the laptop camera pointing up at you.

Anything that alters your image or voice. Do not use portrait mode, beauty filters or digital zoom. Wear plain clothes without fine stripes, which shimmer on video. No earphones, and no glasses that reflect light.

There is no need to perform, either. The twin should look like you in an ordinary meeting, not like you in an advert. Speak at the pace you use with a client.

Two or three, and upload the best. The first usually comes out stiff; the second, more natural. If you slip up during the speaking part, it does not matter: carry on as you would in a meeting.

Before uploading, watch the take through once. Check that you can be heard clearly, that your face is evenly lit and that your hand never covers your mouth.

After you upload it, the recording is checked. The studio confirms the take meets what the twin needs, and the Legathum team supporting your set-up reviews the result before moving on.

Meanwhile, you can move ahead with the part that adds the most value: the brain. Gather your materials (presentations of your method, template proposals, answers to the questions you are asked most) and start writing your decision rules. Face and voice make the twin look like you; the brain makes it useful.

Your image and your voice are yours, and the law protects them. In Spain, Organic Law 1/1982 treats using a person's name, voice or image for commercial purposes without their express consent as an unlawful intrusion, and makes that consent revocable at any time. The GDPR treats physical features that uniquely identify a person, such as facial images, as biometric data and gives them special protection.

That is why Digital Twin by Legathum only clones real people who record themselves. Consent is given by email, with a video of you reading a sentence; the team reviews every twin by hand before it speaks, to prevent impersonation; and you can revoke consent with one click, at which point the twin stops speaking immediately.

Yes, when the twin talks to someone or when you publish a video of it. Article 50 of the EU AI Act requires whoever deploys a system that generates video or audio resembling a real person to disclose that it was artificially generated, clearly and at the latest at the first interaction.

The twin does its part: it says it is an AI in its first turn and its name in the meeting ends in "Clon IA". If you publish videos of it on your website or profiles, add a visible label saying so.

• Today: the roughly seven-minute recording in the studio, with face and voice from the same recording.
• Today: real people only, consent by email with a video reading a sentence, manual review before the twin speaks and revocation with one click.
• In development: the rest of the guided journey inside the studio, which the Legathum team supports today.