Virtual influencers as brand characters
Live influencer or virtual: when each works better
A meta-analysis of 210 experimental studies yielded a result that breaks marketing mythology: virtual influencers trigger more novelty, but noticeably less trust than real people. The audience perceives them as an interesting object, not as a person whose recommendations are worth believing. This is not a verdict on the technology — it is a pointer to where it is appropriate.
The key factor is the type of product. For utilitarian, functional categories (technology, financial services, delivery, B2B-software), a virtual character works more effectively than a real person: the consumer values the function, not the emotional connection with the recommender. For hedonistic categories — fashion, cosmetics, experiences — people still outperform digital characters, except in cases where the character has a high level of anthropomorphism (looks as much like a human as possible) and the audience is already well-versed in the product.
| Criterion | A virtual character is more effective | A live influencer is more effective |
|---|---|---|
| The product type | Utilitarian categories: technology, finance, services | Hedonistic: fashion, beauty, experiences |
| The mechanism of influence | Novelty, emotion, recognizability | Trust, authority, credibility |
| Knowledge of the audience about the product | Works with high knowledge + high anthropomorphism | Works with low product knowledge |
| Reputation risk | Controlled, predictable | Depends on human behavior |
| Cost of scaling content | Low after launch (software, assets) | Growing with every shoot |
A telling example of the right category choice is Demi, the virtual ambassador of the sports brand Demix. The character was trained in movements by real champions in gymnastics, running, football, and dancing over the course of four weeks, after which she was filmed in joint content with Olympic and Paralympic athletes. The product is functional, and the audience is sports-savvy — thus, the digital character here does not compete with a human for trust, but complements the brand's functional story with novelty.
Partnership with a ready-made influencer or creating your own: three models
The choice between 'take a ready-made' and 'create your own' is primarily a question of budget and planning horizon, not creative ambitions. Three fundamentally different models have emerged on the market.
| Model | Cost | Control over the image | When it fits |
|---|---|---|---|
| Build (create your own) | from $250K to $2M+ for development, plus staff and content | Full ownership, the character belongs to the brand forever | Long-term strategy, the character as an element of the brand's DNA |
| License (rent someone else's) | $0 upfront, $20K–$500K per campaign | Limited, the character is used according to a contract | Point-based campaign, hypothesis testing without major investments |
| Generate (generation via software) | from software subscription, hundreds of dollars per asset | Average, depends on the tool | Scaling content for an existing character |
Licensing is suitable for companies that want to test whether the virtual identity format works for their audience before investing hundreds of thousands of dollars. Such was the case with the brands that worked with Lil Mikel — the first truly successful virtual influencer launched in 2016 year: with her, Prada, Calvin Klein, Samsung and BMW collaborated, one post in 2023 year cost $10 000, and the total revenue from the brands exceeded $10 million. But Mikel belongs to the agency Brud, not to her partners — this is an external asset embedded into an external campaign.
Creating one's own character is justified only when the brand is ready to invest in it as a long-term asset, not as a human-faced ad spot. That is exactly what SberBank did with SberKot and M.Video with Em.Vi — both characters fully belong to their respective brands and have been developed over years.
A separate case is AI generation without CGI
And a fourth path has emerged, intermediate between License and Generate: a character entirely created by neural networks without traditional CGI animation. Summer from EPICA became the first such influencer in 2023 year and two months, and independently managed the brand's SMM accounts for two months. Spanish Aitana López earned €10 000 per month already in the second month after launch — this case sparked a wave of interest among marketers in Generate models precisely due to the low entry threshold.
The ominous valley: how to design a character so as not to fall into it
The ominous valley arises not from poor graphics, but from level mismatch. When the face is almost photorealistic, but movements or voice reveal digital nature, the viewer's brain detects the inconsistency and reacts with rejection. Successful brands solve this problem not by improving the render, but by consciously choosing the level of anthropomorphism — and maintaining it consistent across all details.
- Highly anthropomorphic characters (maximally similar to humans) must work through trust and expertise — they need a developed biography, consistent behavior, reaction to real events, not just beautiful appearance.
- Low anthropomorphic characters — mascots, stylized heroes — are freed from the requirement of realism and work through novelty and emotion; attempting to make a mascot 'too human' destroys its function.
- The character concept is chosen before the appearance: first, it is determined what role it plays in the relationships with the audience — expert, friend, provocateur, mascot — and only then is the face drawn.
- Voice, speech rate, and manner of communication in comments should be fixed in the guideline just as strictly as the logo, otherwise the character will 'spread out' between teams and platforms.
- It is useful to test focus groups' reactions specifically on borderline details — eye movement, facial expressions during conversation, rather than static images, because the valley emerges in dynamics.
The transition of SberKot from 2D to 3D in 2023 year is a good example of a conscious, not cosmetic decision. After the rebranding, engagement in advertising content tripled, sales doubled, brand recognizability as innovative among young audience rose by 15 points, and media coverage quintupled. This is the result not of a 'more beautiful' model, but of a consistent low-anthropomorphic mascot, which 3D added expressiveness to without claiming photorealism — and therefore it did not fall into the valley.
Legal and ethical risks: marking is mandatory, not optional
In the past two years, regulation of virtual characters has significantly tightened, and this is no longer a theoretical risk, but an operational requirement affecting the project's economy.
| Jurisdiction | Requirement | What this means for the brand |
|---|---|---|
| China | Mandatory marking with the label 'this is a digital character' | Transparency is built into the law, there is no way around it |
| EU (AI Act, August 2026) | Disclosure of AI-generated content | Marking must be built into content production in advance |
| USA, FTC | Disclosure of material connection to the brand | The character must explicitly indicate advertising posts |
| California, AB 2655 | Labeling deepfakes | Limits the use of hyper-realistic digital identities without marking |
Beyond formal labeling, the company still faces questions that the law has not yet resolved on its behalf: who is responsible if a character 'says' something offensive due to a generative model failure; how to protect the brand image from deepfake fabrications by third-party authors; is it acceptable to present a character's opinion as organic when it is actually a marketing department script. A honest approach — to label even where the law does not yet require it, because the audience reacts more sharply to secrecy than to the mere fact of using a digital avatar.
From campaign to asset: how a character lives longer than the advertising budget
The main mistake is treating a virtual influencer as a creative find for just one campaign. The character becomes a brand asset when it gets its own production infrastructure, not just a one-time video. M. Vi from M. Video received dedicated software Studio.iVFX for generating video content: this reduced editing costs by 3–5 times and enabled the production of Full HD videos in various formats — 16:9, 9:16, selfies, 1:1 — in minutes, not weeks of shooting. It is precisely this production base, not a one-time 'viral' effect, that distinguishes a character-asset from a character-feature.
This practically means embedding the character into the brand's architecture at several levels:
- The character is assigned a role within the brand communication system — it is not just 'a face for advertising', but a channel with its own voice, responsible for a specific part of the user journey (for example, customer support, product training, entertainment content).
- A content production line is created — templates, software, an asset library — so that the cost of producing new materials decreases over time, not increases.
- Metrics are defined that measure not views and likes, but long-term effects: growth in spontaneous brand recognition, changes in perception of brand attributes (innovation, expertise), and audience return to the character's content after months.
- The character is allowed to evolve — to change its image, respond to cultural events, and acquire a history, just like living ambassadors — otherwise, it quickly becomes outdated and turns into a recognizable but burned-out symbol.
It is the increase in brand recognition among youth by 15 points and the doubling of sales after the rebranding of SberKot that is an example of a metric that speaks to the contribution to long-term loyalty, not just a spike in reach. Marginality for the operator of a virtual character at 50–80% versus 20–30% for people-influencers makes such investments economically justified precisely over a few years' horizon — not as a one-time burst of interest in a new technology.
Live influencer or virtual: when each works better
A meta-analysis of 210 experimental studies yielded a result that breaks marketing mythology: virtual influencers trigger more novelty, but noticeably less trust than real people. The audience perceives them as an interesting object, not as a person whose recommendations are worth believing. This is not a verdict on the technology — it is a pointer to where it is appropriate.
The key factor is the type of product. For utilitarian, functional categories (technology, financial services, delivery, B2B-software), a virtual character works more effectively than a real person: the consumer values the function, not the emotional connection with the recommender. For hedonistic categories — fashion, cosmetics, experiences — people still outperform digital characters, except in cases where the character has a high level of anthropomorphism (looks as much like a human as possible) and the audience is already well-versed in the product.
| Criterion | A virtual character is more effective | A live influencer is more effective |
|---|---|---|
| The product type | Utilitarian categories: technology, finance, services | Hedonistic: fashion, beauty, experiences |
| The mechanism of influence | Novelty, emotion, recognizability | Trust, authority, credibility |
| Knowledge of the audience about the product | Works with high knowledge + high anthropomorphism | Works with low product knowledge |
| Reputation risk | Controlled, predictable | Depends on human behavior |
| Cost of scaling content | Low after launch (software, assets) | Growing with every shoot |
A telling example of the right category choice is Demi, the virtual ambassador of the sports brand Demix. The character was trained in movements by real champions in gymnastics, running, football, and dancing over the course of four weeks, after which she was filmed in joint content with Olympic and Paralympic athletes. The product is functional, and the audience is sports-savvy — thus, the digital character here does not compete with a human for trust, but complements the brand's functional story with novelty.
Partnership with a ready-made influencer or creating your own: three models
The choice between 'take a ready-made' and 'create your own' is primarily a question of budget and planning horizon, not creative ambitions. Three fundamentally different models have emerged on the market.
| Model | Cost | Control over the image | When it fits |
|---|---|---|---|
| Build (create your own) | from $250K to $2M+ for development, plus staff and content | Full ownership, the character belongs to the brand forever | Long-term strategy, the character as an element of the brand's DNA |
| License (rent someone else's) | $0 upfront, $20K–$500K per campaign | Limited, the character is used according to a contract | Point-based campaign, hypothesis testing without major investments |
| Generate (generation via software) | from software subscription, hundreds of dollars per asset | Average, depends on the tool | Scaling content for an existing character |
Licensing is suitable for companies that want to test whether the virtual identity format works for their audience before investing hundreds of thousands of dollars. Such was the case with the brands that worked with Lil Mikel — the first truly successful virtual influencer launched in 2016 year: with her, Prada, Calvin Klein, Samsung and BMW collaborated, one post in 2023 year cost $10 000, and the total revenue from the brands exceeded $10 million. But Mikel belongs to the agency Brud, not to her partners — this is an external asset embedded into an external campaign.
Creating one's own character is justified only when the brand is ready to invest in it as a long-term asset, not as a human-faced ad spot. That is exactly what SberBank did with SberKot and M.Video with Em.Vi — both characters fully belong to their respective brands and have been developed over years.
A separate case is AI generation without CGI
And a fourth path has emerged, intermediate between License and Generate: a character entirely created by neural networks without traditional CGI animation. Summer from EPICA became the first such influencer in 2023 year and two months, and independently managed the brand's SMM accounts for two months. Spanish Aitana López earned €10 000 per month already in the second month after launch — this case sparked a wave of interest among marketers in Generate models precisely due to the low entry threshold.
The ominous valley: how to design a character so as not to fall into it
The ominous valley arises not from poor graphics, but from level mismatch. When the face is almost photorealistic, but movements or voice reveal digital nature, the viewer's brain detects the inconsistency and reacts with rejection. Successful brands solve this problem not by improving the render, but by consciously choosing the level of anthropomorphism — and maintaining it consistent across all details.
- Highly anthropomorphic characters (maximally similar to humans) must work through trust and expertise — they need a developed biography, consistent behavior, reaction to real events, not just beautiful appearance.
- Low anthropomorphic characters — mascots, stylized heroes — are freed from the requirement of realism and work through novelty and emotion; attempting to make a mascot 'too human' destroys its function.
- The character concept is chosen before the appearance: first, it is determined what role it plays in the relationships with the audience — expert, friend, provocateur, mascot — and only then is the face drawn.
- Voice, speech rate, and manner of communication in comments should be fixed in the guideline just as strictly as the logo, otherwise the character will 'spread out' between teams and platforms.
- It is useful to test focus groups' reactions specifically on borderline details — eye movement, facial expressions during conversation, rather than static images, because the valley emerges in dynamics.
The transition of SberKot from 2D to 3D in 2023 year is a good example of a conscious, not cosmetic decision. After the rebranding, engagement in advertising content tripled, sales doubled, brand recognizability as innovative among young audience rose by 15 points, and media coverage quintupled. This is the result not of a 'more beautiful' model, but of a consistent low-anthropomorphic mascot, which 3D added expressiveness to without claiming photorealism — and therefore it did not fall into the valley.
Legal and ethical risks: marking is mandatory, not optional
In the past two years, regulation of virtual characters has significantly tightened, and this is no longer a theoretical risk, but an operational requirement affecting the project's economy.
| Jurisdiction | Requirement | What this means for the brand |
|---|---|---|
| China | Mandatory marking with the label 'this is a digital character' | Transparency is built into the law, there is no way around it |
| EU (AI Act, August 2026) | Disclosure of AI-generated content | Marking must be built into content production in advance |
| USA, FTC | Disclosure of material connection to the brand | The character must explicitly indicate advertising posts |
| California, AB 2655 | Labeling deepfakes | Limits the use of hyper-realistic digital identities without marking |
Beyond formal labeling, the company still faces questions that the law has not yet resolved on its behalf: who is responsible if a character 'says' something offensive due to a generative model failure; how to protect the brand image from deepfake fabrications by third-party authors; is it acceptable to present a character's opinion as organic when it is actually a marketing department script. A honest approach — to label even where the law does not yet require it, because the audience reacts more sharply to secrecy than to the mere fact of using a digital avatar.
From campaign to asset: how a character lives longer than the advertising budget
The main mistake is treating a virtual influencer as a creative find for just one campaign. The character becomes a brand asset when it gets its own production infrastructure, not just a one-time video. M. Vi from M. Video received dedicated software Studio.iVFX for generating video content: this reduced editing costs by 3–5 times and enabled the production of Full HD videos in various formats — 16:9, 9:16, selfies, 1:1 — in minutes, not weeks of shooting. It is precisely this production base, not a one-time 'viral' effect, that distinguishes a character-asset from a character-feature.
This practically means embedding the character into the brand's architecture at several levels:
- The character is assigned a role within the brand communication system — it is not just 'a face for advertising', but a channel with its own voice, responsible for a specific part of the user journey (for example, customer support, product training, entertainment content).
- A content production line is created — templates, software, an asset library — so that the cost of producing new materials decreases over time, not increases.
- Metrics are defined that measure not views and likes, but long-term effects: growth in spontaneous brand recognition, changes in perception of brand attributes (innovation, expertise), and audience return to the character's content after months.
- The character is allowed to evolve — to change its image, respond to cultural events, and acquire a history, just like living ambassadors — otherwise, it quickly becomes outdated and turns into a recognizable but burned-out symbol.
It is the increase in brand recognition among youth by 15 points and the doubling of sales after the rebranding of SberKot that is an example of a metric that speaks to the contribution to long-term loyalty, not just a spike in reach. Marginality for the operator of a virtual character at 50–80% versus 20–30% for people-influencers makes such investments economically justified precisely over a few years' horizon — not as a one-time burst of interest in a new technology.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
