Character consistency in AI video, and why documentaries need something different
Character consistency" is the big feature in AI video tools. For factual films the question is subject consistency: the same real person, place and map across the film. Here's how editors handle it.
Every AI video tool now advertises character consistency: the same generated person across scenes, via reference images. It's a real achievement for fiction. For documentaries it answers a question nobody asked, because the people in a documentary are real, and the footage of them already exists.
Consistency in a documentary means something else
- Subject consistency. The same real person recognisable across decades of photographs: you achieve it by choosing images, not generating them.
- Map consistency. One style, one projection habit, the same label typeface, throughout.
- Grade consistency. Archive from different sources matched in contrast, warmth and grain so it reads as one film. See colour grading for documentaries.
- Typographic consistency. One card style, one lower-third style.
- Voice consistency. One narrator, one register, for the whole film and the whole channel.
The generated-character trap
Using reference images to generate "consistent" footage of a real historical figure gives you a consistent fake. History audiences recognise it, and one generated scene poisons the real footage around it. Reserve generated characters for labelled reconstructions and for fiction. See LTX Studio for documentaries.
How editors keep a documentary consistent
- A look applied to the whole timeline: warmth, contrast, grain.
- One map style and one title system, saved as presets.
- A recurring visual motif: an object, a route, a portrait that returns.
- The same narrator, directed the same way.
- Motion rules: the same Ken Burns settings on every still. See the Ken Burns effect.
In templates
That set of rules is most of what an editor's style is, and it's what a Glisse template captures: look, map style, titles, motion settings, sound pattern and narration register. When the AI adapts the template to a new topic, the film is consistent because the rules are the editor's, applied to real footage sourced for the subject.
FAQ
Does character consistency matter for documentaries?
Not in the generated sense. Documentaries need consistency of real subjects, maps, grade, type and voice across the film.
Should I use reference images to generate a historical figure?
No. Generated footage of real people reads as fabrication. Use real photographs and film.
How do I keep archive footage looking consistent?
Match contrast and warmth, apply one grain over everything, and use one map and title style throughout.
Make documentaries with an editor's taste
Glisse turns a real editor's style into a template. Describe your topic, and AI adapts it into a finished long-form film: licensed archive footage, maps, narration and music. Editors earn every time their template is used.
Join the waitlist →