How Virtual Influencers Are Made: A Look Inside the

Lil Miquela’s first Instagram photo went up in April 2016. It looked like an ordinary lifestyle shot. A young woman in soft afternoon light, a little off-center, and a little off in a way you couldn’t quite put your finger on. By 2026 she has 2.6 million followers, a crowd of imitators, and a market around her estimated at $8.3 billion. This piece is about the unglamorous part under all that: how virtual influencers are made. Not only the rendering. The persona, the voice, and the pipeline that turns a sketch into a character your feed believes in.

Step one: the persona, before the polygons

Most teams start by writing. Not modeling. Before anyone opens 3D software, a designer or strategist drafts the character’s bio: where she’s from, what she cares about, what she’s bad at, which jokes she’d tell and which she wouldn’t. Brud’s documents described Miquela as a 19-year-old Brazilian-American with a music habit and a tendency to overshare. Aww Inc., the Tokyo studio behind Imma, worked from a list of fashion references and cultural markers before landing on the pink bob that made her recognizable.

This matters because the persona is what the audience ends up bonding with. A character with no inner life reads like a 3D demo. One with a steady voice, taste, and history can hold attention over hundreds of posts. That’s the difference between a brand mascot and a creator people treat as a peer.

Step two: the 3D modeling pipeline

The classic route from sketch to feed runs through 3D animation software. Artists sculpt the character’s geometry in Blender or Autodesk Maya, clean up the mesh, paint skin and fabric textures, and rig a skeleton so the character can be posed. Then the model goes into a virtual set, gets lit, and is rendered as a still or a short clip. For top-end characters like Miquela or Brazil’s Lu do Magalu, it looks more like a film VFX pipeline than phone-shot content.

Some details get more care than you’d guess. Skin shaders usually use subsurface scattering, so light seems to sink a fraction of a millimeter into the skin, like real skin. Hair is often simulated strand by strand. The eyes get a thin refractive layer over the iris. Whether the uncanny valley opens up or not depends on all of these working together.

Once the model’s done, the team builds a library of reusable expressions and outfits. A virtual influencer posting three times a week can’t start from zero every time. So most of the production effort goes into reusable assets, not one-off renders.

Step three: the AI-driven approach

More and more virtual influencers skip 3D altogether and use generative AI image models. Midjourney, Stable Diffusion, and in-house diffusion pipelines can produce photos of a character from a text prompt, once the character’s been “anchored” with reference images and fine-tuning. Small studios and solo creators can now launch a working virtual persona without building a single 3D asset.

The catch is consistency. A 3D character looks the same from every angle because she is literally the same model. A diffusion-generated character has to be pinned down again for every post. If the workflow drifts, the face changes, followers notice, and the persona falls apart. Most serious AI-driven studios handle this with character LoRAs (fine-tuned adapter weights that keep the face stable across prompts) plus a strict approval queue.

A hybrid is now common. Build the character once in 3D as a baseline reference, then use generative AI for most of the everyday posting. For now, that’s the route that scales best for virtual influencer marketing teams, who need to post often without paying for full CG production on every image.

Step four: voice, motion, and personality

The image is just one layer. Video posts, livestreams, and short clips need movement, and movement is harder than it looks. Motion capture suits and head-mounted face cameras are standard: a human performer drives the character’s body and face in real time, much like a VTuber setup. For pre-rendered clips, animators still use keyframes for fine control, especially for tiny facial expressions.

Voice is another layer. Some characters use voice actors in a studio. Others use AI voice cloning trained on a reference set. The most polished teams keep a short bible of phrases the character does and doesn’t say, so the caption, the spoken voice, and the visuals all feel like the same person.

Personality lives in the writers’ room. Even the best render falls flat if the captions sound like a brand wrote them. Teams that take the persona seriously usually have one or two writers who own the voice and check every post.

Step five: the content engine

Operationally, a finished virtual influencer is a small content studio. A typical week might be two to three feed posts, daily stories, one short video, and constant replies in comments and DMs. Humans handle some of those replies. More and more are handled by conversational AI tuned to the character.

That’s where “virtual influencer” and “AI conversational companion” start to overlap. A character who can talk in real time with thousands of followers at once isn’t just a face in a feed anymore. A lot of what goes into building an AI conversational companion (persona writing, voice consistency, retrieval-augmented dialogue) is now part of the standard virtual influencer stack. That’s especially true for characters whose business depends on ongoing one-to-one engagement instead of passive following.

Why the build matters

People tend to assume how virtual influencers are made is purely a tech question, separate from why anyone engages with them. It’s closer to the opposite. The choices made in production shape the kind of relationship the audience forms.

Build a character with a heavy persona layer, a consistent voice, and an active reply workflow, and you usually get a smaller, tighter audience that feels spoken to. Build one mainly as a rendering exercise and you get a bigger, looser audience that just watches. Audiences can tell the difference even if they can’t name it. For studios that care about keeping an audience for the long haul, persona and voice work has become the most important part of the pipeline. More important than visual fidelity.

The top virtual influencers mostly follow that pattern. The ones that have held attention for years aren’t always the most photorealistic. They’re the ones whose creators kept the writing honest. Technology is the floor. The relationship it supports is the ceiling.

FAQ

How are virtual influencers actually made?
Most come out of one of three pipelines. There’s 3D modeling and animation (Blender or Maya, with motion capture for movement). There’s generative AI (diffusion image models anchored on reference shots and LoRAs). And there’s a hybrid, where a 3D base model anchors a faster AI-driven posting workflow. All three pair the visuals with a persona document, a voice, and a content team.

Do virtual influencers use real people behind them?
Usually, yes. Motion capture performers handle movement in videos, voice actors or AI voice clones handle audio, and human writers do captions and DMs. The character is the public face. Behind it there’s usually a small team.

How long does it take to make a virtual influencer?
A polished 3D character with a full asset library usually takes a small studio several months. An AI-driven character can go live in weeks. Either way, keeping it posting at a believable pace is ongoing work, not a one-time build.

How much does it cost to create a virtual influencer?
It varies a lot. A solo creator using AI tools might spend a few thousand dollars to launch. A studio-built 3D character with motion capture and its own writing team can cost six figures upfront and five figures a month to run. Trade publications and Statista’s virtual influencer market overview follow the wider spending trend, which is shifting toward AI-driven workflows to save money.