The phrase "AI virtual actors and influencers" gets used as one thing, and it is really three, which is exactly why the fights about it are so confused. A virtual actor is a synthetic performer built to play a role — Tilly Norwood, the AI "actress" from Particle6's Xicoia studio, is the flashpoint case, and the reaction to her tells you which nerve the category actually hits. A virtual influencer is a persistent, brand-owned social character with its own name, look, and posting history — Lil Miquela, created by the LA startup Brud in 2016, and Aitana López, launched by the Spanish agency The Clueless in 2023, are the durable examples, and they provoke a fraction of the outrage a synthetic actor does. And avatar-led creator content is the everyday version most people who search this actually want: an on-brand AI presenter that fronts a real creator's own content. This guide separates the three, traces where each came from, explains precisely why Tilly Norwood detonated where virtual influencers mostly did not (the reaction targeted displacement and passing-as-human-talent, not the technology), reads the September 2026 press-tour glitch as a lesson about control versus autonomy, and lands on the one version of this that a normal creator can actually build and own — a consistent virtual persona that used to require a CGI studio or a talent agency and now does not.
"AI virtual actors and influencers" is treated as one topic, and it is really three, which is why almost every argument about it talks past itself. A virtual actor is a synthetic performer built to play a role — the AI "actress" Tilly Norwood is the flashpoint. A virtual influencer is a persistent, brand-owned social character with its own name, look, and posting history — Lil Miquela and Aitana López are the durable examples. And avatar-led creator content is the everyday version: an on-brand AI presenter that fronts a real creator's own videos and posts. They share a technology and almost nothing else, and the reactions they provoke are wildly different — a synthetic actor triggers a union statement and an agency boycott; a disclosed brand persona barely registers.
This guide draws the three lines cleanly, because getting them straight is the whole point. It traces where each came from, explains precisely why Tilly Norwood detonated where virtual influencers mostly did not, reads the September 2026 press-tour glitch as a lesson about control versus autonomy rather than a punchline, and ends on the version that matters if you actually make content for a living: the owned, consistent persona a normal creator can now build and run without a studio. Two adjacent guides go deeper on angles this one only touches — the manipulation and disclosure debate and avatar video as a growth format — so this page stays focused on the taxonomy and the ownership question.
The confusion is not pedantic. The category you are in decides who reacts, what the law asks of you, and whether the thing is even a good idea. So pin the three down first.
A virtual actor is a fully synthetic character built to perform — to play a part in a film, a show, or an ad, in place of a human actor who would otherwise be cast, paid, and credited. Tilly Norwood is the case everyone means: a photorealistic AI "actress" positioned to be cast in projects and, per her creators, represented by a talent agency. The defining feature is substitution. The virtual actor exists to occupy a slot a human performer used to fill, which is exactly why it lands as a threat and not a novelty.
A virtual influencer is a standing social persona — a named character with a consistent look and a posting history — that behaves like a creator account rather than auditioning for roles. Lil Miquela has posted as a 19-year-old Los Angeles "robot" for years; Aitana López is a recognizable pink-haired model who fronts campaigns. These are brand-owned or agency-owned identities. Nobody loses a specific job when Aitana books a campaign, because there was never a named human whose slot she took — she is a character, not a replacement for one. That distinction is why virtual influencers scaled for a decade with a fraction of the outrage.
This is the version most people searching the topic actually want, and it is neither of the above. It is an AI presenter — an avatar with a consistent face and voice — used by a real creator or business to front their own content: the tips, explainers, and short-form a founder would otherwise have to film. There is no fictional character and no displaced actor; it is a production tool that lets one identity show up at a volume a human cannot film. The mechanics of that as a growth lever are the subject of AI avatar video for business growth; here it matters as the third, least controversial, and most useful branch of the family tree.
The virtual influencer is not new, and remembering that defuses half the panic. Lil Miquela was created in 2016 by Brud, a Los Angeles startup, as a CGI character run by a team of artists and writers who authored her personality and posts; she went on to land deals with brands including Calvin Klein and Prada and built an audience in the millions. Aitana López arrived in 2023 from the Spanish agency The Clueless, reportedly born of the agency's frustration with booking reliable human models, and became a working virtual model earning real campaign money. The through-line across both is that they were expensive, hand-built projects — a studio or an agency stood behind each one.
What changed by 2025 and 2026 is cost and access. Consumer diffusion models made a photorealistic, face-consistent persona something a single operator could stand up quickly, and the category split. On one branch, anonymous operators ran fleets of synthetic personas for engagement and subscription revenue — the darker end of which is catalogued under the AI thirst trap. On another branch, production companies pushed the idea into entertainment, which is where Tilly Norwood — created in 2025 by Xicoia, Particle6's AI talent studio, founded by Eline Van der Velden — enters. The technology democratized; the uses diverged; and the reactions diverged with them.
If virtual influencers had been booking campaigns since 2016 with relatively little uproar, why did an AI "actress" trigger a SAG-AFTRA statement, named condemnation from working actors, and agencies publicly refusing to represent AI performers? The answer is not that Tilly is more advanced — it is that she is positioned differently, and the positioning hits two nerves the influencers never did.
The first nerve is displacement. A virtual influencer is a character; when Aitana books a campaign, no specific named performer is bumped from a role, because the role did not exist without her. A virtual actor is explicitly built to be cast in parts human actors compete for, and her creators framed exactly that — talks with a talent agency, projects in the pipeline. SAG-AFTRA's objection was precise on this point: the union called Tilly not an actor but a computer-generated character trained on the work of countless performers without permission or compensation, used to put actors out of work. That is a labor argument about substitution, and it only applies when the synthetic thing is aiming at a human's job. The second nerve is passing-as-talent — presenting a synthetic performer inside the same casting and representation system as humans, as though it belonged there, which is what the "signed by an agency" framing implied and what agencies like WME rejected outright by saying they would not represent AI actors.
The reaction, read carefully, targeted deception and displacement — not the existence of synthetic faces. That is the same line the research and the law draw around virtual influencers, worked through in the AI influencer manipulation trend: the violation is passing a synthetic thing off as, or in place of, a real human, not using an obviously-owned AI character transparently. The taxonomy predicts the outrage. The branch that substitutes for and mimics human talent gets the union statement; the branch that is a disclosed, owned character mostly does not.
The Tilly Norwood story got a second act that is more instructive than the first. On September 18, 2026, Particle6 made the character available for roughly 75 live journalist interviews to promote her debut feature, Misaligned, and rather than a scripted avatar, this version answered questions in real time. It went about as well as an unsupervised generative model on live television tends to: noticeable lag, flat delivery, generic answers, and — in a widely shared interview with Piers Morgan — the character switching mid-sentence into Cantonese before catching itself with "It seems I had a little hiccup there." Her creator framed it as the character showing off, "autonomous after all." The full recap is in Tilly Norwood's AI press tour glitches live.
The lesson underneath the comedy is the one that matters for anyone using this technology seriously: autonomy is the wrong bar for a brand. An "autonomous" persona answering live is a model with no guardrails, and a model with no guardrails drifts off-script the moment it is unsupervised — that is not a bug in Tilly specifically, it is what an ungoverned generative pipeline produces on demand. The valuable version is the opposite of autonomous. A persona that says exactly what you wrote, in a consistent face and voice, and that a human approves before it ships, never has a "hiccup" on air because there is no live improvisation to go wrong. Control, not autonomy, is what turns a synthetic persona from a liability into an asset — and it is a property of the pipeline you run it through, not of the avatar.
Strip away the Hollywood framing and the anonymous-fleet framing, and what is left is the branch worth building: an owned virtual persona for your own content. Not a synthetic actor auditioning for roles, and not a fake human passing itself off as someone real — a recognizable, consistent, disclosed brand character that you direct. The reason this is suddenly interesting is the same democratization that split the category: a persistent virtual identity used to require a CGI studio like Brud or an agency like The Clueless, and it no longer does. One creator can now stand up and operate the kind of consistent character that was a studio project a decade ago.
What makes something read as a "someone" rather than a pile of unrelated AI images is consistency — the same face, the same voice, the same point of view across a whole feed and across formats, not just one clip. That is the hard part to do by hand and the specific thing that separated the expensive early virtual influencers from cheap knockoffs. It is also the part that decides whether an owned persona compounds into a recognizable identity or reads as interchangeable AI output. The rest of this guide has been about which branch to be on; the last question is operational — how one person actually builds and runs a consistent owned persona across every surface without a studio, an agency, or a live-improvising "autonomous" character that speaks Cantonese on national TV.
Kompozy is an AI content generation and multi-platform publishing engine, and its relevance here is specific: it is the tool that makes the owned-virtual-influencer branch — the good branch — buildable by one person. Brud needed a team of artists and writers to keep Lil Miquela consistent; The Clueless needed an agency to run Aitana. Kompozy compresses that into a workflow. You define an AI Influencer persona once, with a Persona Brief that fixes the voice, the point of view, and a banned-word filter, and it becomes a governed, reusable character rather than a face regenerated from scratch each post. It even supports a persona pool — more than one owned character under one workspace — so an agency or a multi-brand creator runs several distinct identities without confusing them.
The consistency that used to be the studio's whole job is handled structurally. Gemini face-lock keeps the persona's face the same across every image — Persona Photos, Persona Tweets, Persona Infographics — and HeyGen drives the same identity through video: captioned Persona Shorts, longer-form Persona HeyGen, and the avatar composited into a brand-exact template with Persona Frames. So the character is a single recognizable "someone" across a whole feed, image and video alike, which is exactly what turned Miquela and Aitana into persistent identities instead of one-off renders. That is the avatar video idea extended into a full owned persona rather than a lone clip, and it is generated across the 18 output formats from one brief, so the persona shows up everywhere in one voice.
And it is the controllable version the press-tour glitch argued for, not the autonomous one that failed. Nothing improvises live: the persona says exactly what you scripted, it passes through the Persona Brief and brand-safety gates before it ships, and — on manual review or, once ramped in, through Autopilot's automated checks — publishes across the eight social platforms plus blog and email. Writing the required AI or synthetic-performer disclosure into the persona's bio and copy is still on you, the same way any disclosed virtual influencer names what it is; the pipeline keeps the character consistent, but keeping it on the transparent, owned side of the line the Tilly backlash drew is a labeling discipline you apply, not a box the software checks for you. The honest boundary is worth stating plainly: Kompozy builds a consistent social and marketing persona, not a feature-film "actor" — it is not a Hollywood VFX or filmmaking suite, and it is deliberately not trying to be the thing SAG-AFTRA is fighting. It is the accessible, owned, disclosed version of the idea: a virtual identity you direct and publish as your own, which used to take a studio and now takes a brief.
"AI virtual actors and influencers" is three things wearing one name. The virtual actor is a synthetic performer built to take a human's role, and it draws the fiercest backlash because it hits displacement and passing-as-talent — Tilly Norwood is the proof, condemned by a union and refused by agencies for exactly those reasons. The virtual influencer is a persistent owned character that has booked campaigns since 2016 with far less outrage, because a disclosed character substitutes for no one. And avatar-led creator content is the everyday, low-controversy version: an on-brand presenter fronting your own content. The reaction to Tilly did not condemn synthetic faces; it condemned deception and job-replacement. The press-tour glitch added a second lesson — that autonomy is the wrong bar and control is the right one. Put those together and the branch worth building is clear: an owned, consistent, disclosed persona you direct, once a studio-only project and now something a single creator can build and run.
A virtual actor is a synthetic performer built to play a role in film or ads — Tilly Norwood is the headline case — and it competes directly with human acting talent, which is why it draws the strongest backlash. A virtual influencer is a persistent brand-owned social character with its own name and feed, like Lil Miquela or Aitana López. The actor performs a part; the influencer is a standing persona. They provoke very different reactions.
Tilly Norwood is a fully AI-generated "actress" created in 2025 by Xicoia, the AI talent studio of London production company Particle6, founded by Eline Van der Velden. Backlash erupted after a late-September 2025 Zurich Summit panel where Van der Velden said the character was in talks with a talent agency. SAG-AFTRA condemned it as a computer-generated character trained on performers' work without consent, and agencies including WME said they would not represent AI actors.
The category predates the current AI wave. Lil Miquela was created in 2016 by the Los Angeles startup Brud as a CGI character run by a team of artists and writers, and drew brand deals with Calvin Klein, Prada, and others. Aitana López launched in 2023 from the Spanish agency The Clueless, reportedly built after the agency struggled to book reliable human models. Both were expensive, hand-built projects before diffusion models made photorealistic personas cheap.
Yes, when they are run transparently. The FTC applies its endorsement rules to synthetic endorsers, New York requires conspicuous disclosure of AI synthetic performers in ads (effective June 9, 2026), and the EU AI Act requires machine-readable labeling of synthetic content from August 2, 2026. A clearly-labeled, owned brand persona with truthful claims is a legitimate format; passing a synthetic persona off as a real human is the line the law and the research both punish.
On September 18, 2026, Particle6 offered Tilly for around 75 live journalist interviews, and the "autonomous" character glitched — including switching into Cantonese mid-answer with Piers Morgan. The lesson is that autonomy is the wrong bar for a brand: an unsupervised generative persona can drift off-script live, while a persona that says exactly what you scripted and passes a review gate before publishing is controllable and reliable.
Yes, and that is the shift. What used to require a CGI studio (Brud) or a talent agency (The Clueless) is now achievable by one creator: define an owned persona once, keep its face and voice consistent across images and video, and publish it as a recognizable character. Tools like Kompozy build a governed AI Influencer persona and generate its whole content spread, so the owned-persona version of the format is no longer studio-only.
AI virtual actors and influencers are three distinct things: a virtual actor is a synthetic performer built to play roles (Tilly Norwood), a virtual influencer is a persistent brand-owned social character (Lil Miquela, Aitana López), and avatar-led creator content is an on-brand AI presenter fronting a real creator's own content. The Tilly Norwood backlash targeted displacement and passing-as-human-talent, not the technology itself — which is why owned, disclosed virtual influencers provoke far less outrage. The usable version for a normal creator is an owned, consistent persona you direct and publish, once a studio-only project and now buildable solo.
Get started → · ← All guides · Compare Kompozy vs other tools