A character is not one reference image
A portrait captures one moment. A reusable character asset also needs age, proportions, facial structure, hair, wardrobe layers, props, movement habits and features that must survive different framings.
Three layers of a visual anchor
A useful anchor answers identity, styling and world questions at the same time.
- Identity: what cannot change
- Styling: what belongs to this scene
- World: what rules govern camera, material, color and light
Write attributes the camera can verify
Prefer observable attributes over literary adjectives. Instead of only 'restrained', describe an avoidance of eye contact or a pause before turning—details that can enter performance direction.
Approve before scaling
Batch generation multiplies a bad premise. Validate the character and visual language with a small number of images, then reference the approved version in later shots.
Create a new version deliberately
A text edit does not always require a new portrait. Generate a new visual version when age, hair, silhouette, core wardrobe or recognition features change, and keep older versions available for shots that already depend on them.