The AI influencer
glossary.
Every word this category uses, defined once and in plain English — from AI influencer and virtual influencer to the working vocabulary of seeds, LoRAs, outfit swaps and disclosure labels.
Half these words
mean the same thing.
The vocabulary around generated people arrived in layers — a 3D era, a diffusion era, now a platform-policy era — and each layer left its own name for roughly the same idea. Several entries below are interchangeable in ordinary use, and where that is true the definition says so instead of inventing a distinction.
Terms run A to Z and each has its own anchor, so a single definition can be linked or quoted on its own. Where a concept has a full page behind it, the entry points there. Where a term is genuinely contested — UGC, deepfake, CGI — you get the argument rather than a side.
20 terms.
One page.
Persona, format, craft, commerce and policy words, alphabetically. Pick one to jump to it.
The terms,
defined.
Two to four sentences each. Neutral where the industry is neutral, honest where it is not.
- Craft
AI avatar
A generated face used as an on-camera presenter. The word describes the surface rather than the method: a rigged 3D head and a diffusion-generated person both qualify, though it now usually means the second. An avatar is a face; an influencer is a face with an audience.
- Persona
AI influencer
A social identity whose personality and posting schedule are real but whose face is generated. The term says nothing about the pipeline — diffusion, 3D, or a mix — only that nobody was photographed. It is the most searched name for the category and, partly because of that, the loosest.
- Format
AI UGC
Creator-style content — talk to camera, product in hand — made with a generated actor and a script instead of a filmed creator. It borrows the format of UGC, not its provenance, which is why people who read “user-generated” literally object to the name. Widely used anyway.
- Commerce
Brand ambassador
A face contracted to represent a brand over time rather than for a single campaign. With a generated character the arrangement becomes ownership instead of a contract: no renewal, no availability window, no defection to a competitor. The advertising obligations do not change.
- Persona
CGI influencer
A virtual persona built the traditional way: modelled, rigged, textured, lit and rendered in 3D. The label names a production pipeline, and it belongs to the mid-2010s characters that started the category. It is now applied loosely to any non-real person, including ones with no 3D in them.
- Craft
Consistency
How reliably a character's face and body stay the same across separate generations. It is the central technical problem of the format: image models sample fresh every time, so identity has to be pinned with a trained character model, reference images, or both. The failure mode is drift.
- Policy
Deepfake
Synthetic media that places a real, identifiable person in something they did not do or say. The line between this and an AI influencer is consent: a deepfake reuses an existing likeness, a character is invented. The word gets stretched to cover any AI video, which blurs a distinction worth keeping.
- Persona
Digital human
Usually a real-time, interactive avatar — a brand presenter, a support agent, a kiosk face — rendered live and able to respond. Vendors use it for speech-driven characters, so it optimises for conversation rather than for a feed. It meets AI influencer only where one character does both.
- Policy
Disclosure (AI labelling)
Telling the audience that content is AI-generated. TikTok and Meta both provide labels and expect realistic synthetic content to carry one, and advertising law on misleading claims applies whoever is on camera. Requirements are set per platform and change often, so check the current policy.
- Format
Faceless creator
An account run without showing the operator's face: voiceover, b-roll, text on screen, or a generated presenter. It is how many people arrive at AI influencers — the aim is publishing without being personally identifiable, not passing the character off as human.
- Commerce
Lookbook
A set of images presenting a collection as complete styled outfits, traditionally shot on one model so the clothes stay the variable. Generated lookbooks invert the constraint: the model is free and constant, so a whole season can share one face without rebooking anyone.
- Craft
LoRA / character model
A small file trained on top of a larger image model to teach it one specific subject — here, one face. LoRA is short for Low-Rank Adaptation; “character model” is the same idea in plainer words. Loading it makes a person reusable across new prompts.
- Craft
Outfit swap
Changing what a character wears while holding the face, pose and framing. Distinct from virtual try-on, which starts from a real garment and aims at a faithful fit. A swap is a styling operation, and it does not by itself promise the garment is reproduced exactly.
- Commerce
Product placement
Putting a real product into a generated scene so the character holds, wears or uses it. The craft is matching label, proportion and lighting to the rest of the frame. Accuracy matters beyond aesthetics: a product that misrepresents the real item is an advertising problem.
- Craft
Prompt
The written instruction describing the image or clip you want. Prompts specify scene, wardrobe, lighting and camera well, and identity badly — a description of a face returns a slightly different person every time it is sampled. That gap is the reason character models exist.
- Craft
Seed
The number that initialises the random noise a diffusion model starts from. The same seed with the same prompt and settings reproduces the same image; change it and you get a variation. It is a reproducibility control, not an identity control.
- Persona
Synthetic influencer
A near-synonym for AI influencer, borrowed from “synthetic media”. Press, analysts and policy documents prefer it; the people running the accounts almost never use it about themselves. No technical distinction separates the two terms, so use whichever one your reader already uses.
- Policy
Synthetic media
The umbrella term for images, audio and video generated or substantially altered by a model rather than captured by a camera. It is the vocabulary platforms and regulators write rules in, which is why it appears in labelling policy far more often than in marketing copy. Deepfakes and AI influencers are both subsets.
- Format
UGC
User-generated content: media made by customers and ordinary creators rather than by a brand's production team. In practice it names a look as much as a source — one person, one phone, one room. That is what makes it imitable, and “AI UGC” coherent despite the contradiction.
- Persona
Virtual influencer
The older name for the same category, in general use before generative image models arrived and still standard in press and academic writing. It covers CGI characters and generated ones equally, because it describes what the persona is rather than how it is made.
Where each word
gets its own page.
A definition tells you what a word means, not what the work involves. AI influencer covers building and running one. Synthetic creator traces where the persona words came from. AI UGC takes the format apart, including the jobs it should not be given.
The AI model generator is where consistency and character models stop being vocabulary and become settings. Fashion is where lookbooks, outfit swaps and product placement earn their keep, and Feedroll answers the prompt entry above: concepts that already worked.
Keep exploring
Now build
the influencer.
Pick a trained face from the Marketplace or upload your own model. Everything defined on this page is a button once you are inside.
AI influencer included · Cancel anytime · Commercial use