Person Profiles and the Face Wedge

What is a Person?

A Person is a face profile you create by uploading 1–5 reference photos of the same subject. Once created, you can attach it to any generation and the AI will render that face consistently across every thumbnail.

Navigate to Persons in the sidebar to create and manage your profiles.

The face wedge

The wedge is the quality mechanism that makes face consistency reliable. It works by:

1. Putting the Person first in the prompt. The generation engine places the FOREGROUND identity-lock block — the person's face and descriptions — before any style guidance or inspiration references. This ordering prevents a style-reference image from being read as the subject.

2. Framing inspirations as mood-board only. Any Style Reference you attach is explicitly labelled "for mood and composition only, not a face to reproduce." This stops an inspiration image that happens to contain a face from overriding the locked identity.

3. Competing-face detection. If a Style Reference appears to contain a face (detected from its caption and style analysis), the editor shows a warning badge. The face in the style ref is suppressed in the prompt so it cannot compete with your Person.

Adding a Person to a generation

In the Advanced lane (Studio), open the Person panel and select the profile you want. In the Quick lane, the Person selector is available on the generate screen.

Only one Person is used per generation. If you want two people (host + guest), use Podcast Mode — see the Podcast Thumbnails guide.

Strict Likeness toggle

The Prioritize looking like me toggle (strictLikeness) is on by default. When enabled, the prompt instructs the model to reproduce the face exactly — same expression, same features. Turning it off gives the model more creative latitude with the face, useful when you want a stylized or illustrated look rather than a photorealistic one.

Best practices for face photos

• Upload 3–5 varied photos (different angles, expressions, lighting conditions).
• Use photos where the face is forward-facing, clearly visible, and well-lit.
• Avoid group shots or photos where another face is prominent — the AI validation rejects uploads that contain multiple clear faces or where no single face is dominant.
• Higher-quality source photos produce more consistent outputs.

Avatar validation

When you upload a face photo, FatThumb runs three checks before accepting it:

1. Forward-facing — the face must be roughly facing the camera.
2. Single face — the image must contain exactly one dominant face.
3. Eyes visible — both eyes must be discernible.

Photos that fail validation are rejected with a specific reason so you can re-upload a better image.