Presenters — Use Your Own People On Camera

Alfred and Ada ship as standard presenters: name either in a prompt and the clip is generated from their official look, so their face is the same every time.

You are not limited to those two. Drop a few photos — or one short clip — into the presenter folder and anyone can be cast the same way: by name, in plain English, with the same consistent face across every video.

Register someone in three steps

1. Find the presenter folder

Presenters live in one folder of their own, so nobody's likeness is stored anywhere but your own workspace:

<your user data folder>/_presenters

Set PRESENTER_ANCHORS_DIR to put it somewhere else. Create the folder if it isn't there yet.

2. Add their anchors

An anchor is a reference picture of the person. Name each file <slug>_<role>:

_presenters/
  mia_front.png      ← required: the face the clip opens on
  mia_hero.jpg       ← optional: a second strong front angle
  mia_profile.png    ← optional: side view
  mia_back.mp4       ← optional: rear view

• slug — lowercase letters, numbers and hyphens (mia, jo-patel). This is their handle.
• role — one of front, hero, profile, back. front and hero come first; one of them becomes the opening frame of the clip.
• format — png, jpg, jpeg, webp, or a video: mp4, mov, m4v, webm.

One front anchor is enough to get started. More angles give Alfrada more to work from, so the person holds together better when they turn or move.

A clip works as well as a portrait

Point a phone at someone for two seconds and save it as mia_front.mp4. One representative frame is picked out automatically — skipping black lead-in, blinks and motion blur — and used as the anchor. It is the quickest way to register somebody who hasn't got a good studio photo.

3. Give them a name and a description

Add a presenters.json in the same folder:

{
  "mia": {
    "name": "Mia Chen",
    "notes": "Late twenties, dark curly hair, charcoal blazer over a white tee. Warm, quick, direct."
  },
  "jo-patel": {
    "name": "Jo Patel",
    "notes": "Grey crew-neck, short beard, round glasses. Dry and unhurried."
  }
}

• name is what you call them in a prompt. Without this the slug is title-cased (jo-patel → "Jo Patel").
• notes describe their manner and clothing — not their face. The face comes from the anchors; notes keep the wardrobe and mood steady between clips.

That's it. The folder is read fresh each time, so a new person is castable immediately — no restart.

Cast them

Name them exactly as you would Alfred or Ada:

"Make an 8-second clip of Mia looking into the camera. She says: Here's what changed this quarter."

"Put that clip over in/demo.mp4 in Presenter Mode, Mia in a box bottom-right."

The rules that apply to Alfred and Ada apply to everyone:

• Describe what they say and do, never how they look. The anchors settle their appearance. Describing a face in the prompt fights the anchors and you get a different person.
• Put the words after says:. Everything from says: up to the first stage direction is what they say; a direction that follows, such as "She smiles and points at the chart.", is performed, not spoken. One says: per line of speech; for a back-and-forth, make one clip per line and use the Script lane below. Keep a prompt under about 600 characters so the whole brief reaches the generator.
• Ask for the length the words need. About two words a second plus a second of lead-in: six words is a 4-second clip, twelve words an 8-second one. A longer request is trimmed to that fit, because unused seconds get filled with extra speech.
• One name per clip. If a prompt names two registered presenters, neither is applied automatically — pass presenter explicitly to pick one.
• A short all-caps name is read as an acronym. ADA means the statute, not Ada. The same guard protects any registered name of three letters or fewer.
• Opt out with presenter set to none for an ordinary Video Lab generation.

Two presenters in one video

A video does not have to be one person talking for its whole length. The Script lane in Live → Video Studio holds a sequence of lines, and each line says who is on camera and what they say — so Alfred can open, Ada can take the numbers, and anyone you have registered can close.

Each line carries three things:

• Presenter — the slug of whoever is speaking (alfred, ada, mia).
• Says — their line, kept with the clip so the script stays readable and editable without opening the video.
• Clip — the generated clip that plays for that line.

Drag a line's right edge to change how long it is on screen; the lines after it move with it, so no silent gap opens up. Delete a line and the rest close up behind it. The preview switches presenter exactly where the export does.

Lines are output time

A line's start and end are positions in the finished video, the same as layout segments and freezes. A clip shorter than the span it is given holds its last frame for the remainder rather than dragging everything after it earlier.

Character sheets are handled for you

If your reference is a character sheet — a figure inside a flat border with a caption strip underneath — the panel is cropped out automatically so only the person is used. You do not need to crop it yourself.

Anchors are also framed to the clip's aspect ratio on the way in. For a tall source the top is kept, so the head stays in shot.

Permission matters

A presenter is a recognisable person. Register someone only when you have their agreement to be generated on camera and to have those clips used the way you intend. This is someone's likeness and voice-alike, not stock footage — "they work here" is not consent, and a photo you happen to have is not permission.

Keep the folder to people who have actually said yes, and remove them when that changes: deleting their files de-registers them immediately.

Troubleshooting

What you see	What it means
The name isn't recognised	The filename doesn't match <slug>_<role>.<ext>, or the role isn't one of front/hero/profile/back. Check the spelling and the extension.
A different person is generated	The prompt described a face. Remove the appearance wording and let the anchors do it.
A video anchor is ignored	No frame could be read from that clip. Try a different clip, or add a still image instead.
The clip is cropped oddly	The anchor is very tall or very wide. Add a front anchor framed roughly head-and-shoulders.
Both names in one prompt, neither applied	Two registered presenters were named. Pass the one you want explicitly.

See also

• Presenter Mode release notes — putting a presenter over a screen recording
• Live Workspace — the Video Studio these clips land in