Skip to main content
Two tools from the “UGC” family put real looking people on camera: “Talking avatar” (one person alone) and “AI podcast” (two hosts at most). The video model itself says each line, with the voice fixed at the first speaking shot.

​
Talking avatar

A photo of you or a generated person says your text, filmed like on a phone, lips synced. The field is called “What you write”, with two choices:
  • “My script”: said word for word.
  • “An idea”: the AI writes the text, the way people talk to camera.

​
The form that works

“My script” mode:
Write the way you talk: short sentences, one filler word (“well”, “honestly”), a real pause. Count about 20 words for 15 seconds.
“An idea” mode: give the topic and the tone, the AI writes the text.

​
Before / after

No precise topic, no tone, no audience.

​
Avatar settings

  • Limit: 1,200 characters. The length is deduced from the text (about 2 words per second, between 10 s and 4 minutes). In idea mode, you choose the length.
  • Photo: optional. A sharp, front facing, well lit face is used as is. Without a photo, a person is generated; “Consistent character” keeps their face the same from one shot to the next.
  • Voice: “Automatic” creates the voice at the first shot and keeps it afterward. Pick a voice to impose it.
  • Avatar model and Resolution: they set the quality of the render and the price per second.
  • Subtitles: on by default. Music: off by default.
  • A speaking shot lasts 15 s at most: a long text is cut into several shots.

​
AI podcast

A two voice dialogue, one host per shot, full screen or split screen: the editing alternates between full screen on the one speaking and a vertical split screen, one host on top, the other at the bottom, where the listener reacts.

​
The form that works

Start each line with the host’s name in square brackets, like [Alex]. Two hosts at most.
No script at hand? Write a topic of a few words (at least 3) in “Generate a dialogue with AI”: the dialogue is written into the field and you review it before launching. It is a billed text call.

​
Before / after

No brackets, no topic, lines with no substance.

​
Podcast settings

  • Host photos (optional): one photo per host, named as in the dialogue. Without a photo, the host is generated; “Consistent hosts” keeps their face the same from one shot to the next.
  • Limit: 1,200 characters, default length 30 s. Subtitles on, music off.
  • Each host has a single image and a single voice, reused on all their lines.

​
Common mistakes

  • Avatar: a text written to be read: long sentences sound fake. Cut and talk.
  • Avatar: a group photo or a profile shot: give a sharp, front facing face.
  • Podcast: three hosts or more: two at most.
  • Podcast: a different name on each line: keep exactly the same names: they are used to recognize each host and their photos.
  • Going over 1,200 characters: the field stops. Shorten, or make several videos.
See also Paste your script and the voice.