Same face in every scene: Mia in a cafe, on a rooftop and in the rain

Diffusion models don't remember your hero. Here's what causes character drift in AI video, what doesn't fix it, and the actor-based workflow that does.

How this video was made

Generated in Actoria from a three-paragraph script: each paragraph became a five-second scene, the synthetic actors were created once and kept their faces in every shot, and the scenes were stitched into one film.

Scenes

1. 0s Mia sits at a window table in a small daytime cafe, copper hair catching the light, olive green bomber jacket on. She stirs her coffee, looks up at the window and smiles at something outside. Slow push-in on her face.
2. 5s Night. Mia stands on a rooftop above a glowing city, the same olive green bomber jacket zipped up against the wind. She turns toward the camera as neon signs flicker far below. Handheld, slight drift.
3. 10s Rainy street at dusk, reflections on wet asphalt. Mia walks toward the camera under a clear umbrella, olive green jacket, copper hair tucked behind one ear, and stops with a confident grin. Tracking shot backing away in front of her.