Identity lock is the feature that keeps one person's face the same across every AI image or video you generate. Without it, each generation invents a slightly different face; with it, the output stays recognisably you.
This guide defines identity lock, compares it with other approaches, and shows where it matters: video series, multi-platform photos and UGC at scale.
Related: identity lock prompts for AI photos and videos.
- Identity lock is face (and overall likeness) consistency across AI outputs: the model reads a real person's face from a reference photo and holds it fixed.
- Pose AI does this with Nano Banana 2 for images, from one selfie and with no training step, and carries the same identity into Kling and SeedDance video.
- It matters most for series, profiles on several platforms, and UGC where the same creator must appear again and again.
- It applies to your own face or someone who has agreed to it. It does not make a stranger's face reusable.
What is identity lock?
Identity lock is a generation method in which an AI model takes a person's face from a reference photo and keeps it fixed while everything else changes: scene, outfit, light, angle or motion. The prompt controls the surroundings; the photo controls the person.
It is different from describing a face in words, which can match millions of faces, and different from training a custom model, which needs many photos and a wait.
Identity lock vs other approaches
| Approach | How the face is chosen | Consistency | Effort |
|---|---|---|---|
| Identity lock (Pose) | Read from one reference photo at generation time | High, across images and video | Low: one selfie, no training |
| Multi-face or prompt-only generation | Described in words; a new face each time | Low; drifts between images | High, and still unreliable |
| Single-shot tools | One image per run, no memory of the last | None across a set | Medium: repeat and hope |
| Custom model training | Fine-tuned on a batch of photos | High after training | High: many photos and a wait |
If you need the same person across more than one image, the choice is between identity lock and training a model. Identity lock is the faster of the two.
Where identity lock matters
Video series: the person on screen in episode one has to be the person in episode ten, and a locked identity carries from still images into video clips. Multi-platform photos: a LinkedIn headshot, an Instagram shot and a dating profile photo should read as the same person. UGC at scale: a creator persona repeated across many ad variants keeps an audience's recognition.
Pose applies this in Image Studio and carries it into Kling and SeedDance in the Video Studio.
Use it responsibly
Identity lock reproduces a real face. Use your own photo or one you have clear permission to use, and follow platform rules about disclosing AI-generated content.
Try it with AI headshots.
See it in motion in the Pose AI video studio.
Browse professional headshot styles.
Writing prompts for a locked identity: identity lock prompts.