Identity Collapse

Identity Collapse refers to what happens when an AI video generator depicts the same character across multiple scenes, but the face subtly changes each time — eventually looking like someone else entirely compared to the original.

Think of asking a painter to draw a person from a verbal description alone, with no photo, and having them redraw that person for each new scene. The first painting might look close enough, but by the tenth painting using the same description, the eyes or face shape have drifted just enough that the result is barely recognizable. AI video models run into the same problem because they often regenerate a character's face from scratch every time a scene changes.

The fix is to create one reference image of the character's face ahead of time, then feed that same reference image into every subsequent scene. Instead of re-describing the character each time, repeatedly calling back to a fixed reference image keeps the face stable.

Articles use the phrase "the 'identity collapse' problem that commonly occurs when inserting a real person's face into AI-generated video." It's easy to confuse this with deepfake technology, which deliberately swaps faces, but it's actually the opposite — this term describes the unwanted side effect of a face drifting while the AI is trying to keep drawing the same person.

Before generating a video, create a single reference image — a character sheet showing the person's face from the front and side. Then attach this character sheet as a reference image to every scene prompt you write afterward.

Example prompt: "Generate [scene description] with the same face as the person in this character sheet. The facial features and face shape must match the reference image exactly."