Glossary

Key terms used across AutoPersonas and AI-influencer operations. For deeper detail, see the documentation.

AI influencer: A synthetic social-media persona with a fixed face, body, styling, and voice, operated to publish content and engage an audience across platforms. Unlike a one-off AI-generated portrait, an AI influencer is persistent: the same identity shows up across hundreds of posts, months of publishing, and multiple networks, accumulating followers the way a human creator does. Example: a fashion persona posting a daily outfit-of-the-day to X and Fanvue, replying to comments in its own voice, and carrying brand partnerships — while remaining recognizably the same person from post 1 to post 1,000. See the full guide to AI influencers for history, notable examples, and business models.
Persona: The structured identity of an AI influencer: appearance, personality, wardrobe, aesthetic palette, and writing voice. In AutoPersonas a persona is more than a system prompt — it's a structured record with typed fields (age, build, ethnicity, subculture, signifiers, named wardrobe sets, photography register) that drives every image, caption, and reply the influencer produces. Example: a persona defined as "31, lean build, Mexican-Japanese, outdoorsy dog-dad, paw-print wrist tattoo" renders the same recognizable man whether the scene is a trail hike or a coffee run. Structuring identity this way, rather than re-describing it per prompt, is what prevents drift; the prompt-engineering guide has the field-by-field breakdown.
Profile: The publishing identity layer in AutoPersonas. Every content item, social account connection, and analytics snapshot is scoped to a profile. A profile may be backed by a generated AI influencer or run as a managed (real) account — the publishing, scheduling, and analytics machinery is identical either way. Example: an agency running five AI influencers plus one real brand account operates six profiles from one dashboard, each with its own queue, connected platforms, and engagement stats. Profiles are also what plan tiers count: the Free plan includes one influencer, Pro includes five.
Reference sheet: A multi-view composite image — front, side, back, and a face close-up — that locks an AI influencer's identity so its appearance stays consistent across thousands of generated images and videos. The front panel is the identity anchor; the other panels are derived from it so every angle agrees on the same face, hairline, and build. Example: without a reference sheet, a prompt like "27-year-old fashion blogger" produces a different woman on every run; with one, a three-quarter-angle café shot and a full-body street shot render the same person. The reference sheet is the core input to consistent character generation.
Identity lock: Holding an AI influencer's face, body, wardrobe, and aesthetic steady across every generation, platform, and month of operation — the core problem AutoPersonas solves. Identity lock combines a reference sheet, named wardrobe sets, a fixed photography register, and post-generation face-match QA that rejects renders drifting below a similarity threshold. Example: a viewer scrolling back three months through a locked persona's feed sees one person; scrolling an unlocked feed, they see plausible siblings — and that single moment of doubt is what kills accounts. Read the visual consistency problem for why single-image consistency was solved but operational consistency wasn't.
Managed account: A real brand or person's account operated through AutoPersonas' publishing surface — captioning, scheduling, review queue, and analytics — without generating an AI influencer at all. Managed accounts exist because half the pipeline (automated social media content) is useful even when the face in the photos is real. Example: a boutique owner connects her existing X account, uploads her own product photos, and lets the platform draft captions in her established voice and publish on schedule while she approves posts from the queue. Details in the managed accounts docs.
LoRA: Low-Rank Adaptation — a lightweight fine-tuning technique that trains a small adapter on an AI influencer's reference images to further improve likeness in image generation, without retraining the underlying model. A LoRA is trained once per influencer (typically a few dollars of compute) and then amortizes across thousands of renders, which is why it features in every honest cost breakdown. Example: after a persona's reference sheet is locked, training a face LoRA on those views tightens fine facial geometry — the details that drift first — in difficult shots like low light or extreme profile angles.
Voice lock: Keeping captions and replies consistent with an AI influencer's established tone and personality, so written output reads unmistakably as the same person post after post. Voice lock is defined with structured fields — formality level, humor register, emoji policy, signature phrases, and a banned-vocabulary list of LLM tells — rather than a vague "write casually" instruction. Example: a persona locked to "lowercase, dry, no exclamation points, occasional parentheticals" produces "biscuit found a tennis ball under the porch this morning (we have eleven already)" instead of generic AI copy like "These quiet moments are pure magic." The prompt-engineering guide covers caption-side prompting in depth.
Engagement engine: Automated, voice-consistent replies and outbound discovery that grow and maintain an AI influencer's audience without manual work. The engagement engine watches mentions and comments on connected platforms, drafts responses in the persona's locked voice, and can proactively engage with relevant accounts in the niche. Example: when a follower comments on a fitness influencer's post asking about the workout, the engine replies with a specific, in-character answer within minutes — the reply cadence platform algorithms reward and solo operators can't sustain. Engagement automation is included in the Pro plan; docs at engagement engines.
Learning loop: The continuous-learning pipeline that watches which posts land — engagement, reach — and biases future generations toward what performed, by tuning per-tag and per-time weights that feed back into content generation. Example: if a travel influencer's golden-hour vista posts consistently outperform its food shots, the loop gradually raises the scene planner's weighting for vista-style scenes and shifts publishing toward the times its audience actually engages. Over months this compounds into a feed shaped by real audience data rather than the operator's guesses. The loop's inputs come from the same data surfaced in analytics & insights.
Cross-persona analytics: Roster-level analytics that aggregate engagement across every AI influencer an operator runs, with a per-persona breakdown — distinct from single-profile stats. Example: an agency running five personas sees total reach and engagement for the roster, then drills into the fact that the beauty persona drives 60% of engagement on 30% of the posts, informing where next month's generation budget goes. For agencies running rosters at scale this is the primary reporting surface; see analytics & insights.
A/B testing: Defining control and test content variants for an AI influencer, measuring each arm's real engagement, and declaring a winner by lift once both arms reach a minimum sample size. Example: run ten posts captioned in the persona's terse register against ten in a warmer register; if the warm arm shows a meaningful lift in engagement per impression, it wins and future captions adopt it. Because generation is cheap and identity is locked, AI influencers can A/B test at a cadence human creators can't — the face stays constant while one variable at a time changes.
Usage-based billing: AutoPersonas pricing has two layers: a subscription tier and metered usage on top. Tiers: Free at $0 (one influencer, one connected account, pay-per-use with a card on file), Pro at $50/month or $500/year (five influencers, video generation, engagement automation, Flow Engine beta), and Agency at custom pricing. Usage is billed per operation regardless of tier: $0.01 per published post, ~$0.21 per image render, and text at $0.45 per million input tokens / $3.75 per million output tokens. Example: a Pro subscriber posting daily typically sees $30-50/month of usage on top of the $50 subscription. See pricing and billing & plans.
MCP (Model Context Protocol): An open protocol that lets AI agents call tools on a server. AutoPersonas exposes its full product surface over a hosted MCP server at mcp.autopersonas.com, so agents can create AI influencers, generate and review content, and schedule posts programmatically — the same operations as the dashboard, over a protocol agents natively speak. Example: an operator connects the AutoPersonas MCP server to their agent of choice and asks it to "draft this week's posts for my fitness persona and queue them for review," and the agent orchestrates it end to end through MCP tools. Setup guide: MCP docs; REST alternative: developer API.
Collaboration studio: Tooling for multi-influencer collaborations and branded shoots, where two or more AI influencers appear together in the same generated scenes — consistently. The hard problem is holding two locked identities in one frame without either drifting; the studio conditions generation on both reference sheets simultaneously. Example: a fashion persona and a travel persona "meet up" for a city-break campaign, each recognizably themselves across the shared posts, cross-tagging each other so both audiences see the collab. Docs: collaboration studio.
Flow Engine: AutoPersonas' visual automation builder, currently in beta and available on Pro and above. Flows chain triggers, conditions, and actions into repeatable pipelines that run without operator attention — the orchestration layer above individual posts. Example: a flow that triggers every weekday at 7am, generates three candidate posts for a persona, runs them through moderation, publishes the best-scoring one to X and Fanvue, and drops the rest into the review queue for manual salvage. Docs: flows.
AI Assistant: A chat interface inside AutoPersonas that can operate the whole platform conversationally. Available on any plan with a payment method attached, billed pay-per-token rather than by subscription tier. The assistant can create and edit personas, draft and schedule content, inspect analytics, and explain what it changed. Example: typing "my beauty persona's engagement dipped last week — figure out why and adjust this week's plan" produces an analysis of the recent posts plus a proposed, executable content plan. Docs: AI assistant.
Orchestration: Running a roster of AI influencers as a coordinated operation rather than a pile of individual accounts: shared scheduling, per-persona queues, cross-persona analytics, engagement automation, and automation flows under one control surface. Orchestration is the framing behind AutoPersonas' positioning as an AI influencer orchestration platform — the generation problem is largely solved; the operational problem of keeping five personas posting, replying, and improving simultaneously is what the platform automates. Example: one operator running five niched personas ships 35 posts a week with under an hour a day of review time.
Video generation: Producing short-form video clips of an AI influencer — image-to-video renders that preserve the persona's locked identity in motion. Video is included in the Pro plan and billed per second of rendered output, with practical costs dominated by takes: most published clips are the best of two or three renders. Example: a 6-second clip of a fitness persona demonstrating a stretch, generated from a locked still so the face survives the motion, published as a native video post. For the cost math, see the cost breakdown.
Moderation queue: The pre-publish review surface where every generated post lands before it can ship. Content is automatically screened — face-match QA against the reference sheet, content-safety checks, caption-quality gates — and the operator approves, edits, or rejects from the queue. Example: of five renders generated for a post slot, two pass automated checks and reach the queue; the operator keeps one, and the rejection data feeds the learning loop. The queue is why realistic cost budgeting assumes roughly five generations per two published posts. Docs: content safety.