AI АВАТАРЫ: КАК ЭТО РАБОТАЕТ? Деньги на ИИ-Моделях

Business-focused summary: how “AI avatars” are monetized and operated

Core concept & positioning

• The speakers describe neuroavatars (AI-generated fictional characters) as a new content-and-monetization industry.
• Monetization is driven less by “magic tools” and more by:
  • archetype / legend
  • funnel execution
  • testing
• Avatars can be used for:
  • Direct paid content (subscriptions, donations, custom requests)
  • UGC-style funnels and influencer development (using generated characters as “content engines”)

---

Monetization model (how cash is made)

Primary revenue streams mentioned

• Selling avatar content (described as the main cash driver for the primary presenter)
• Customizations / custom content, including:
  • “dance in underwear”
  • “say my name”
  • “humiliate me” / “video-signature” (videosigna style)
  • chatting-related custom interactions
• Chatting (presented as a major ecosystem driver)
  • Claim: chatting can represent ~70% of total earnings in “normal companies/systems”
  • The presenter’s own split is said to be more content-sales oriented
• Community monetization
  • boosts / patrons / tributes
  • subscriptions, donations, and record-based payments (“stars” system)

Concrete performance examples & KPIs (as stated)

• Early performance (new account / initial ramp)
  • First month: 150k “stars” (≈ 150 rubles mentioned as a rough equivalence)
  • Another donor metric: 442,747 stars for Don, converted to about 447,000 rubles (as stated in subtitles)
• Top creators snapshot
  • A creator (e.g., “Lily Bennet”) reportedly earned ~1,000 rubles (subtitle numbers appear confusing, but presented as proof of monetization)
  • Recorded donations: ~30,000 stars for a creator
• Speed-to-traction case study (Telegram funnel)
  • Million views by the 5th uploaded video
  • 10,000 subscribers in one week
  • After launching a Telegram channel: 2–3 days to generate early-paid sales
  • Reported: ~$130 “this morning” for a few days after Telegram launch (used as evidence of scaling potential)

Revenue targets/timelines (explicit)

• Personal target: ~2 million rubles per month
• Time target: ~6–7 hours/day
• Teaching/offer target: “course from zero to monetization” (timeline implied as relatively short, not strictly specified)
• Turnkey production target:
  • Deliverables: 40 pieces of content (30 videos + 10 photos, per subtitles)
  • Output timeframe: ~1 month
  • Generation time for the team: ~2 hours/day
  • Performance claim for client: ~120k in a month (stated after turnkey delivery)

---

Funnel & growth playbook (process)

Funnel architecture (described in words)

• Build an avatar:
  • archetype
  • appearance
  • legend
• Post content to drive discovery:
  • recommended: Threads first with “trigger posts” to seed attention and funnel traffic
• Convert interest into monetized channels:
  • Telegram links and “stars”/donation systems
  • use customizations and limited “warming up” to drive paid upgrades

“Warm-up” sequencing (actionable recommendation)

• Operational rule: don’t immediately sell the most explicit content.
• Example logic:
  • If interest becomes “perfectly clear” too fast (too explicit too quickly), demand drops.
• Demonstrated with a case:
  • Later explicit monetization existed, but the account was later banned, ending that approach’s audience.

Speed vs. tooling (operational principle)

• The presenter argues against overly complex “workflow choreography” (a long step-by-step pipeline).
• Instead:
  • “Do it quickly.”
  • Test hypotheses fast through content iteration rather than perfecting technical pipelines.

---

Strategy: archetype/legend as the main lever

Framework-like takeaways (explicitly emphasized)

• “First need a legend; beginners make the mistake of focusing only on a beautiful girl.”
• The “story” drives conversion.
• Similar-looking avatars can still fail without differentiation.

Two development paths for a legend

• Develop archetype first, then attach appearance
• Or (for visually driven builders): create appearance first, then attach:
  • hobbies/interests
  • backstory (e.g., “why is she in a wheelchair?”, “why a scar?”)
  • emotional/identity hooks
    • example includes bullying → therapy → public strength narrative

Example narrative hook (used as a model)

• Backstory example:
  • a character has a scar
  • the video includes a trauma origin (bitten/bullied at age 6 → therapy → “strong & independent” arc)
• Used to illustrate creating “Black Mirror”-style motivation.

---

Testing & iteration (core operating cadence)

• Constant A/B testing through rapid generation:
  • entrepreneurs generate products/content quickly and watch whether people ask, “I want to order this.”
• Traffic-to-sales conversion is non-linear:
  • large subscriber counts don’t guarantee revenue
  • failure mode: poor conversion (“didn’t process traffic well”)

---

Team & scalability playbook

Team roles (stated)

• Current team: 2 people (presenter + wife)
  • delegation examples:
    • wife generates/creates new content
    • voice/motion control tasks (also easier continuity with her voice/movements)

Scaling options

• If scaling:
  • add chatters once traffic grows (human-in-the-loop conversation support)
  • move to an agency-like model:
    • outsource generation to a production specialist
    • outsource community management / “chatters” for throughput

Practical team target

• For “avatar creation only”: 2 people enough
• Additional experts are needed later for courses and broader use cases

---

Tools & tech stack (operational)

Mentioned neural networks/platform components

• NanoBan: generating avatar appearance (images)
• Cdream 4.5: generating NSFW content (noted as for “strawberries”)
• CE 2.5: another generation model used for motion/creation (“top is CE 2.5” in subtitles)
• Clean Motion Control:
  • described as less relevant/outdated
  • motion tracking issues mentioned (e.g., “face swimming”)

Prompting approach

• Prompts can be very long: ~24 pages long for a video
• High detail reduces artifacts, but errors still happen (e.g., hallucinations, physical inconsistencies)

Actionable operational risk/reliability note

• Prompt errors can force regeneration:
  • e.g., wrong movements (somersaults rarely physically correct)
• Regeneration increases cost.

---

Distribution & regulation risk management

Social platform recommendations (execution)

• Best-performing approach (per presenter’s experience):
  • Start where posting is simpler and cheaper for generation-to-post workflows
  • “No air-slinging” recommended (avoid features that may trigger moderation behavior)
• Platform guidance (as stated):
  • Telegram: often works well for links/lead capture
  • Threads: “top right now” with a cheap posting workflow
  • Instagram: possible but higher moderation risk; requires more spend to sustain output
  • Facebook/VK: presenter reports personal bans on Facebook; uncertainty about VK
• Explicit risk:
  • accounts may be kicked/banned even when AI tagging/official policies exist

Monetization survival plan (account risk)

• The presenter’s account was banned (mentions April and proof recorded).
• Plan:
  • restore the account
  • avoid showing new models in some situations due to toxic reporting behavior

Legal/disputes note (high level)

• Risks exist (reports, legal costs), but the presenter reports no major personal legal-cost incidents.
• Complaints are described as mostly content similarity/identity disputes rather than full legal cases.
• Key risk management takeaway:
  • “haters/schoolchildren” flooding reports can “ruin the whole thing,” even if isolated cases don’t.

---

Market outlook (high-level, execution-focused)

• Claim: avatars will take increasing market share from live bloggers because:
  • easier to scale content production
  • less human churn (no human factor)
• Also: businesses will use AI-generated assets for rapid A/B testing of products before hiring full teams.

---

All presenters / sources

• Yuri Mikhailovich Kusto (main presenter/source)
• Co-presenter: unnamed interviewer/host (not identified in subtitles; referred to informally as “Mr. Kilgare” / “Jack” in parts)