The curated collection of photos (typically 8–20) used to train a persona. Reference quality and variety — angles, lighting, expressions — directly determine how consistent and robust the trained identity is.
A good reference set is a small, deliberate portfolio of one identity: front, three-quarter and profile angles; indoor and outdoor light; neutral and expressive faces; head-and-shoulders plus at least a few wider crops that show build. The model can only generalize what it has seen — variety in the set becomes flexibility in the persona.
The common mistakes are all forms of sameness or noise: a burst of near-identical selfies from one session, heavy beauty filters (the model learns the filter, not the face), sunglasses and occlusions, other people in frame, and low-resolution or compressed images. Any of these quietly caps the quality of every future generation.
There is also a rights dimension: reference photos must not depict a real, identifiable third party you have no rights to. InfluencerForge.app personas are original synthetic identities — the brand-safe persona guide explains how to build a clean set, and the training guide shows how the set feeds into training.
Related terms
The one-time process of teaching an image-generation model a specific persona's identity from a set of reference photos. After training, the persona can be generated in unlimited new scenes without re-describing its appearance in each prompt.
Fixing a persona's facial identity at training time so that every subsequent generation shares the same face and build. The alternative — re-describing a character in each prompt — produces visible identity drift between images.
A reference document that fixes everything about a persona beyond the trained face: style rules, wardrobe boundaries, voice and tone, backstory, and explicit do's and don'ts. Teams use it to keep the persona consistent across people and campaigns.