Skip to content

Chapter 3 — Virtual Influencer

👤

"We created Aitana because we were tired of real models' drama.3 years later, she's the busiest model in Spain."Rubén Cruz, The Clueless Agency, Barcelona

You'll learn

  • 3 virtual influencer business models (sponsorship, ambassador, subscription)
  • Consistency stack: LoRA + IPAdapter + ControlNet → 80-90% similarity
  • Case studies: Aitana, Lil Miquela, Vi An (Vietnam)
  • A-to-Z pipeline for creating an AI persona
  • Legal + ethics: voice cloning + AI disclosure

01 Aitana López — €10K/month in Barcelona

Background

The Clueless is a creative agency in Barcelona. Founders: Rubén Cruz + Diana Núñez.

They were tired of real model drama (late, demanding more, doesn't fit brand) → created a hyper-real AI model.

Result: Aitana López — Spanish 25-year-old, pink hair, fitness body.

Numbers

MetricNumber
IG followers (peak)~4.3M (currently stable ~378K)
Average revenue€3,000/month
Peak€10,000/month
Brand ambassadorBig supplements (Spain)
Brand collabZara, Sephora
SubscriptionFanvue (NSFW tier)
Sister modelMaia (also from The Clueless)

The Clueless stack

StepTool
Base modelStable Diffusion 1.5 (later Flux)
Identity LoRATrain custom LoRA for Aitana
CompositionIPAdapter
PoseControlNet OpenPose
PolishPhotoshop manual touch-up
PostingWeekly content meeting, 4-5 posts/week

Industry-defining quote

"We are mainly taken aback by the skyrocketing costs of those [real] influencers."Diana Núñez


02 Lil Miquela — pioneer virtual influencer

Profile (US reference)

ItemNumber
AgencyBrud (Los Angeles)
Launch2016
IG followers~2.6M
Rate / post~$10,000
Annual revenue~$10M
Brand listCalvin Klein, Prada, Samsung

Lesson from Brud: virtual influencers don't need hyper-realism to succeed. Lil Miquela is clearly 3D-animated — but "personality" + cohesive content built the moat.


03 Vi An — Vietnam's first AI ambassador

Background

ADT Creative (HCMC, founded 2015) spent 3 years developing Vi An. Goal: Vietnam's first hyper-real virtual brand ambassador.

"Vi An" = "Việt" + "An" (peace).

Numbers

ItemNumber
Development time3 years
StackCGI + 3D scan + motion capture + AI Human
EngineUnreal Engine + Houdini + Maya
Main brandViettel — Y-Fest 2024 ambassador
DebutMWC 2024
IG followers (@vian.righthere)~38K
Celeb collaborationsAnh Tú, Diệu Nhi, Tun Phạm

Lesson

"Vietnam doesn't need to wait for an Aitana from the US.One major brand (Viettel) is enough to fund a 3-year project + dedicated agency.Another 100 Vietnamese brands (Vinamilk, Vingroup, Highlands, Trung Nguyên...) don't yet have virtual ambassadors."


04 7-step pipeline to create a hyper-real AI persona

Standard 2026 workflow

1. Persona ──→ 2. Visual ──→ 3. LoRA ──→ 4. Test ──→ 5. IPAdapter ──→ 6. Post ──→ 7. Monetize
   (concept)    (10 ref pics) (train)     (iterate)   (compose)       (cadence)    (revenue)
   1 day        3 days         2 days     2 days      ongoing         weekly       monthly

Step 1. Persona concept

Define:

  • Name + bio (age, occupation, hobbies, hometown)
  • Personality (fun, deep, quirky...)
  • Niche (fashion, fitness, tech, lifestyle, gaming)
  • Origin story (why this persona exists)
  • Aesthetic (mood board 20+ images ref)

Step 2. Visual identity — 10 reference images

Generate 10 images same person, 10 different angles:

  • Frontal, 3/4, profile (3 basic angles)
  • Close-up, medium, full-body (3 distances)
  • Smiling, neutral, contemplative (3 expressions)
  • Plus 1 signature pose

Tool: Midjourney V7 (--cref chain) or Flux + IPAdapter.

Step 3. Train LoRA

This is the most important step. LoRA = "small fine-tune" to teach the model your character.

PlatformCostSpeed
Replicate (Flux LoRA)~$2-5 / train20-40 min
Fal.ai~$3 / train15-30 min
CivitAIFree (community quota)1-2 hrs
Local (RTX 4090)Free + electricity2-4 hrs

Dataset: 20-50 images from step 2 (diverse angle/expression).

Step 4. Test consistency

Generate 50-100 test images with new LoRA → measure:

  • % similarity to original
  • % recognizable as same person

Target: >80% consistency.

Step 5. IPAdapter + ControlNet (production)

ComfyUI workflow:

  • LoRA Aitana (identity)
  • IPAdapter (style — outfit, location)
  • ControlNet OpenPose (precise pose)
  • ControlNet Depth (composition)

→ Each production image = 1-2 minutes render on consumer GPU.

Step 6. Posting cadence

PlatformRecommended frequency
Instagram4-5 posts/week + 7-10 stories/day
TikTok1-3 videos/day
Twitter/X3-5 posts/day
Fanvue / OnlyFans5-10 posts/week

Step 7. Monetize (detail in section 06)


05 Monetization — 3 streams

3 revenue models

1. Sponsorship / brand deal

  • Aitana: €3K-10K/month from supplements + fashion
  • Rate: $50-500 / 1K followers (depends on niche)

2. Brand ambassador (long-term)

  • Vi An × Viettel = enterprise template
  • Contract 6-12 months
  • Rate: $5K-50K/contract depending on brand size

3. Subscription content (Fanvue / OnlyFans)

  • Aitana has Fanvue tier
  • Rate: $5-30/month/subscriber
  • Highest if niche is fitness / fashion / NSFW

06 Prompt pack — persona creation

5 templates

1. Define persona (Claude / ChatGPT)

Create virtual influencer persona for [target market]:
- Name: [name]
- Age: 22-28
- Occupation: [local-friendly job]
- Niche: [fashion/lifestyle/tech]
- Personality: 3 words
- Aesthetic mood: 5 words
- Origin story: 100 words
- 10 sample captions for IG

2. Visual identity (Midjourney V7)

beautiful [ethnicity] woman, [age], [hair: long black wavy], 
[skin tone], [features: defined cheekbone, almond eyes], 
[outfit: contemporary fashion], 
[lighting: golden hour urban], cinematic, 8K, --ar 4:5 --v 7 --cref [URL]

3. LoRA training dataset prompts 20 captions for 20 dataset images:

1. portrait of [persona], front facing, neutral expression
2. portrait of [persona], 3/4 angle, slight smile
... (20 variations)

4. Production scene (Flux + LoRA)

photo of [persona name], wearing [outfit], at [location], 
[activity], candid, natural light, shot on iPhone 16 Pro

5. Caption (Claude)

Write 5 IG captions for [photo description] of persona [name].
Style: [casual / aspirational / quirky]
Hashtags: #[niche] #[location]
Length: 2-3 lines + hashtag

07 Common pitfalls

🚨 6 mistakes

1. Skip persona definition → beautiful images but "soulless," no fanbase

2. Dataset not diverse → LoRA over-fits, fails on new poses

3. Clone celebs / real people → lawsuits (Disney 2025, Spotify voice clone ban)

4. No AI disclosure → backlash when discovered (Velvet Sundown case)

5. Spam posting → IG/TikTok algo demotes if low engagement

6. Inconsistent personality → each caption a different voice, audience confused


08 Opportunity in emerging markets

5 niches with traction

NicheAudienceBrand fit
Local fashion KOL18-30 femaleLocal fashion, cafés, lifestyle
Tech reviewer20-35 maleSmartphone, gadget, fintech
Food vlogger local cuisine18-40 all regionsF&B, snacks, delivery apps
Travel content (Bali, Hoi An, Phuket)25-45Resort, airline, tour
Mom influencer kids25-40 femaleBaby products, toys, edu

Economics

ItemCost
Midjourney V7 Standard$30/month
Flux + Replicate LoRA$20/month
ComfyUI (local)$0 (needs GPU)
ElevenLabs (for video voice)$5/month
Photoshop subscription$10/month (Photography plan)
Total~$65/month

→ 1 brand deal $500 = 7 months of runway.

  • Must disclose AI when collab'ing with brands
  • Ambassador contract via company (registered business)
  • Tax: brand deal income → declare
  • Avoid cloning real faces (any nationality)

09 Practice exercises

✍️ 3 levels

Level 1 — 1 week

  • Define 1 persona (full bio + mood board)
  • Generate 10 identity images with MJ --cref
  • Post to test IG account

Level 2 — 1 month

  • Train LoRA for persona
  • Post 20 images over 4 weeks, consistency >80%
  • Measure follower growth + engagement

Level 3 — 6 months

  • Build audience 5K-10K real followers
  • Land 1 small brand deal ($200-1K)
  • Test Fanvue tier (if niche fits)

10 Continue reading

Final word

"Aitana isn't real, but her story is. Brands need stories, not people.Vi An proves: 1 Vietnamese brand was willing to pay 3 years + billions for an AI ambassador.You don't need permission. You need persona + consistency + cadence."