What is a virtual influencer?
We read the files 20 AI images carry. All 20 say "Created using Generative AI" before anyone posts them.

A virtual influencer is a computer-made character that posts on social media like a person, with a face, a name and a voice, while a real team writes, designs and decides everything she does. Early ones were 3D renders. Many new ones are made with AI image and video models. Either way, she is not a real person, and US ad rules treat her endorsements like a human's.
Create your AI modelA virtual influencer is a character that does not exist, posting on Instagram, TikTok or YouTube as if she did, while real people run every part of her account. The first wave were 3D renders built by studios. Many new ones are made with AI image and video models from one fixed face, which is the job building an AI influencer step by step walks through. What separates a working virtual influencer from a pile of AI pictures is that her face stays the same in every post, and keeping the same face in every AI photo shows how hard that is without a reference.
Virtual, AI and CGI: the three names
The words get mixed up, so here is how they are used in practice.
- Virtual influencer is the umbrella term: any influencer who is not a real person. It says nothing about the tools.
- CGI influencer means a virtual influencer built as a 3D model, posed and lit like a character in a film or a video game, then placed into photos.
- AI influencer means a virtual influencer whose photos and clips come from generative AI models, built from a saved reference face.
The best-known early case shows the line. Lil Miquela had built an Instagram following when CNN reported in 2018 that she was a computer-generated character made by Brud, a Los Angeles startup. She was CGI, not generative AI, but she was a virtual influencer all the same. Newer characters follow the same idea with different tools: a team picks a face, writes a personality and a niche, and posts as her.
So does "virtual influencer" mean AI? Not by definition. "Virtual" says she is not real. "AI" names the tool. An AI influencer is one kind of virtual influencer.
Our test
Two things set an AI virtual influencer apart from a person with a camera, and we can measure both on our own files: whether the face holds, and what the file itself says about where it came from.
First, the face. We sent one scene prompt, with no face words, to five image models four times each: Nano Banana Pro, GPT Image 2, Seedream 4.5, Wan 2.7 Pro and FLUX.1 dev. The scene came back right in all 20 images. Our character's face came back in none of them. Four models drew four different women each. Seedream 4.5 drew one steady face in all four tries, but it was the model's own default face, not ours. Full results and the image grid are in the same-face test.
That is the whole craft of a virtual influencer in one result. Without a saved face, an AI model gives you a new stranger every time. A CGI team solves this with a 3D model that never changes. An AI team solves it with a reference: in PixelPig, a persona built from one clear face photo, or a trained LoRA from 30 to 40 photos on Pro and Scale.
Second, the file. We read the metadata in the same 20 images with c2pa-python. Every one carried a signed C2PA record marked with the IPTC source type trainedAlgorithmicMedia, which IPTC defines as "Created using Generative AI". The signer differed by model: Google, OpenAI, Byteplus, and fal for the two open models we ran through it. Only the four Nano Banana Pro files also wrote the AI field a second time in plain photo metadata (details in our Instagram label test). The method and the signer table are in what C2PA is and why AI images carry it.
What this means for a virtual influencer: an AI-made photo can arrive at Instagram already marked as AI before the creator types a caption. Meta reads those fields to add its "AI info" label, as our guide to whether Instagram detects AI images explains. A virtual influencer who plans to hide what she is starts from a file that says otherwise.
How a virtual influencer is made today
The steps are the same whichever tool you use. Here is the AI route as it runs in PixelPig.
- Choose a face you have the right to use. On the AI persona page there are two routes: pick a ready-made AI model, or use your own photos of a real, consenting adult. Ready-made models are shared: once you take one she disappears from your own picker, but other accounts can still choose the same face. If the face is the brand, build from photos you own.
- Write the character. Name, age range, voice, what she posts and for whom. The AI influencer niche ideas list helps here. A character with no niche has nothing to say.
- Generate from the persona, not from words. Describe the scene, outfit and light in the prompt and let the persona carry the face. Check each batch against one master image.
- Add motion. Short clips from a still of the persona keep video and photos matching.
- Label it. Use the platform's AI label where the rules ask for it, and disclose paid deals. More on both below.
A CGI team swaps step 3 for 3D modelling, rigging and rendering. That gives total control of the face, but every post needs skilled 3D work, which is why AI is the faster route for a new account.
CGI vs AI vs human: side by side
| CGI virtual influencer | AI virtual influencer | Human influencer | |
|---|---|---|---|
| How the images are made | 3D model, rendered and composited | Image and video models from a reference face | Camera |
| What holds the face | The 3D model itself | A persona reference or a trained LoRA | The person |
| What the file says | Depends on the 3D and editing software | In our test, 20 of 20 carried a signed "Created using Generative AI" record | Camera data, if the phone keeps it |
| Who decides what she posts | A studio or brand team | A creator, agency or brand | The person |
| US ad disclosure for paid posts | Yes | Yes | Yes |
| Platform AI label | When the media is realistic, per platform rules | When the media is realistic, per platform rules | Only for realistic AI-made or heavily AI-edited media |
The rules virtual influencers follow
On ads: in the US, the FTC's Endorsement Guides define an endorser as someone who is, or appears to be, an individual, group or institution (16 CFR 255.0). A character who does not exist but appears to be a person fits that wording. A sponsored post by a character who does not exist needs the same clear "ad" disclosure as a post by a person, and the brand and the people running her can both be on the hook. The label tools and contract questions for a paid post are in our guide to AI influencer brand deals.
On platforms: Meta requires creators to disclose photorealistic AI video and realistic AI audio (Meta), and adds its own label when a file carries the signals it reads. TikTok requires a label on AI content that shows realistic people or scenes (TikTok) and reads Content Credentials too. The specifics, with sources, are in our guide to TikTok's AI content rules.
On likeness: a virtual influencer cannot borrow a real person's face. Our acceptable use policy bans using an identifiable person's likeness without their documented permission, and that includes building a "virtual" version of someone you follow.
Common mistakes
- Treating "virtual" as "anonymous". Real people run the account and answer for it. Brands ask who is behind a character before they sign anything, and teams that run several AI models as an agency keep a written brief and consent record per character for that reason.
- Changing the face between posts. A new face reads as a new person, and the account loses the one thing it is built on. Fix the face in a persona first, then experiment with scenes.
- Building on a face you do not control. A shared ready-made face can show up on another account. Plan for that, or build from photos you own.
- Skipping the label. The file may already say "AI". A missing label on a realistic post risks a platform penalty and reads as a cover-up.
- No niche. A pretty face posting at random gives people no reason to follow. Pick what she is about before the first post.
- Copying a real person. Building a lookalike of a celebrity or a creator you follow breaks our rules and can break the law.
If you want the character without showing your own face at all, the faceless AI creator guide covers that setup and the YouTube, TikTok and Meta rules that come with it.
Bottom line
A virtual influencer is a made-up character with a real team behind her. CGI started the idea, and AI models now do most of the work. Two things decide whether one works: a face that never changes, which takes a saved reference rather than a better prompt, and honest labels, because the AI files we tested already announce themselves as AI. Build the face first, write the character second, and disclose both the AI and the ads.
Questions
- Does virtual influencer mean AI?
- Not always. "Virtual influencer" means the character is not a real person. The first famous ones were 3D renders made by artists. Many new ones are made with AI image and video models, so the two terms often get used for the same thing.
- Are virtual influencers real people?
- No. The character does not exist. Real people run the account, though. A creator, a studio or a brand writes the captions, picks the photos, answers deals and is responsible for what she posts.
- Who makes money from a virtual influencer?
- Whoever owns and runs the character, which is a creator, an agency or a brand. The character signs nothing and earns nothing herself. We do not quote earnings figures, because they depend on the account and the deal.
- Do virtual influencers have to say they are not real?
- The rules ask for disclosure of two things. Paid deals must be disclosed under the FTC Endorsement Guides, which cover endorsers who appear to be a person. Realistic AI media must be labelled under Meta's and TikTok's own rules. Our Instagram and TikTok guides cover the label steps.
- What is the difference between a virtual influencer and an AI influencer?
- An AI influencer is a virtual influencer made with generative AI. A CGI influencer is one built as a 3D model and rendered like a game or film character. Followers see the same thing, a person who does not exist, so marketers use the names for each other.
- How do you make a virtual influencer look the same in every post?
- Fix the face in one reference, then describe only the scene. In our test, a scene prompt with no reference gave 20 photos and none showed our character's face. A PixelPig persona holds the face, and a trained LoRA holds it tighter.
- Can anyone start a virtual influencer?
- Yes, with a face you have the right to use. In PixelPig that means a ready-made AI model or photos of a consenting adult. Using a real person's likeness without permission is against our acceptable use policy and, in many places, the law.