A cast defined once per project and carried across every scene and every video, so a character-driven channel is recognisable at a glance instead of subtly different on each cut.
Generating "an older woman in a red coat" twice gives you two different people. A character is defined once for the project with its own reference images, and every scene uses them — so the fortieth scene is the same person as the first.
Autotube reads each scene of the narration and works out who is actually in it, then generates that scene with only those characters' references attached. A scene with two leads gets both.
A scene with nobody in it stays a scene, rather than having a stand-in inserted to fill the frame.
Character work depends on how many reference images a model accepts and how faithfully it uses them, and the image models differ sharply here. Each model's page says how large a cast it can take.
Every plan includes the full pipeline. The app shows the exact credit cost before you run anything.