Lesson 2 — Generative Video vs AI Avatars
Generative models invent shots that cannot be filmed. Avatars deliver lines. Confuse the two and you burn a week.
- Generative video is for b-roll, transformation and impossible shots; avatars are for talking-head, spokesperson and multi-language delivery.
- Default to roughly 90 percent b-roll and 10 percent talking head, then let performance move the ratio.
- Aleph-class editing models transform footage you already own, turning one hero shoot into many variants without a reshoot.
- Start from a locked still and animate it: image-to-video is far more consistent than text-to-video.
These are two different machines. Generative video invents shots that never existed: b-roll, transformations, impossible camera moves. Avatars deliver a script to camera as a spokesperson, in many languages. Picking wrong burns days.
A strong default mix is roughly ninety percent b-roll and ten percent talking head. The face earns trust for a few seconds; the b-roll does the actual selling. Start there, then let performance data move the ratio, not your taste.
Runway Aleph and tools like it edit existing footage instead of generating from nothing. Recolor a product, swap the background, change the camera angle. One hero shoot becomes twenty market variants with no crew and no reshoot.
For consistency, start from a locked still. Generate the frame in an image model, iterate there where it is cheap and fast, then animate it. Text to video reinvents your product and your lighting on every single run.
Six slots: subject, action, scene, camera move, style, and audio. Then state the aspect ratio explicitly, nine by sixteen for vertical. After a near miss, change exactly one variable. Rewriting everything turns this into a slot machine.
Action step. Take yesterday's shot list and tag every line as generative or avatar. Then build one hero still in an image model, animate it, and do a single reroll where the only thing you change is the camera move.
Loading the quiz…