Lore

Soul ID: Personal Likeness Training from Reference Photos

Overview

Higgsfield's "Soul ID" trains a personal likeness model from roughly 20 varied reference photos of the creator, letting the creator's own face/likeness be inserted into any generated scene afterward.

Method

• Supply ~20 reference photos with varied angle, expression, and lighting (not a single headshot) as training input.
• Once trained, the resulting identity can be dropped into arbitrary scene generations, functioning like a reusable "cast member" the creator plays themselves.

Related

• Distinct from Higgsfield @-Element Asset System, which invokes pre-made assets via @-mentions in prompts — Soul ID instead trains a new likeness from scratch off personal photos.
• Complements Three-Panel Character Sheet for locking a character's appearance once the identity exists.

Из тем: Unsorted, Character and Asset Consistency Across a Project