Context
This project was developed as part of the KISOME program. The goal: build a faceless, ethical, sustainable content brand operating on a 10-hour/week budget — generating quiz and cozy aesthetic content across TikTok, Instagram, and Facebook using AI tools.
Tools used: ChatGPT · FlowithAI / Manus · Midjourney · ElevenLabs · Adobe Premiere · Suno · n8n
Constraints: 10 h/week · 60–80 €/month · 100% anonymity
Project Overview
Style reference: Hiroi Sekai YouTube channel
Two-channel strategy:
- Quiz / Playful Learning — curiosity-driven short-form content, retention-optimised
- Cozy Aesthetic / Aesthetic Enjoyment — mood-regulation content, high save rate
Mission: Create a digital sanctuary — calm, ethical, slow content that builds a loyal community without relying on a personal face or brand identity.
Research Foundation
Psychology & Algorithms
- Curiosity Gap → high watch time on quiz videos
- Mood regulation & gentle fascination → high save rate on cozy content
Platform strengths:
- TikTok → discovery / reach building
- Instagram → aesthetic authority
- Facebook → community reach
Prompt Engineering Workflow
- Build mega-prompts and refine with ChatGPT (5.1 Thinking)
- Two-part structure: algorithm & psychology analysis + personal social media roadmap
- Export as PDF → feed into an agentic LLM (FlowithAI / Manus)
Strategic Blueprint
Long-term goal: Build an ethical, faceless media asset brand
Guardrails: 10 h/week · minimal cost · privacy-first
Target audiences & brand persona:
| Segment | Age | Need |
|---|---|---|
| The Learner | 18–30 | Competence, quick wins |
| The Professional | 30+ | Calm, nostalgia, recovery |
Persona: Calm older sibling figure — gentle, encouraging, never shaming
Platform formats:
- TikTok → retention-oriented puzzles
- Instagram → save-worthy aesthetics
- Facebook → repurposing & discussions
Content Architecture
Cozy Pillars
| Pillar | Description |
|---|---|
| Warm in the Cold (core) | Cozy interior + cold outside world — rain, wind, snow, fog |
| Historic Cosiness | Old libraries, vintage cafés, trains, historic streets |
| Golden Hour Dreams | Warm sunlight, late afternoon, coastal rooms, balconies |
| Mood Blends | Cozy + melancholic / nostalgic / mysterious — always real/historic settings |
| Product-ready Environments | Highly detailed worlds where cups, books, lamps, posters can later be swapped for brand placements |
Quiz Pillars
| Pillar | Description |
|---|---|
| Strange but True | 1 question + 4 answers → reveal → short explanation + bonus fact |
| Unexpected Topics | History → Chemistry → Cinema → Everyday — unpredictable jumps |
| Difficulty Mix | 20% easy · 60% medium · 20% hard |
| Calm Competence Tests | Clear, neutral questions — no countdown stress, no "genius-only" framing |
| Mini Explainers | Last 5–10 seconds: a micro-lesson on the answer |
Production Workflow & Measurement
Weekly 10-hour system:
| Day | Task |
|---|---|
| Mon | Ideas & scripts (8–10 quiz entries, 4–5 cozy prompts) |
| Tue | Asset generation (voiceovers, AI visuals) |
| Wed | Batch editing: cut 6 quiz videos |
| Thu | Scheduling & captions |
| Fri | Community replies & review |
Season system: Each season = 4–6 weeks · stable format → adjustments only after season review
Primary metrics:
- Quiz: retention, completion rate, comments
- Cozy: saves, loops, mood keywords
- Avoid vanity metrics (likes, follower count in early phase)
Adaptation: Monthly reviews (not daily) · identify top-performing topics · soft A/B testing on tone, hook type, pacing
Cozy Style Concept — Visual Reference
Ultra-realistic AI-generated interior — warm lighting, layered textures, plants, and books against a cold city backdrop. This visual tone defines the cozy channel's aesthetic: immersive, slow, and emotionally safe.
Updated Cozy Concept — YouTube Lounge Radio
New content format: 24/7-style lounge radio videos on YouTube — relaxing evening lounge music generated with Suno, seamlessly looped cozy videos.
Visual composition:
- Foreground: cozy interior elements (lamp, table, cup, plants, books) — warm colour palette
- Mid / window surface: transparent projection layer for viewer messages
- Background: city at evening or night — warm window lights, soft glow, subtle movement (rain, fog, clouds)
Monetisation mechanic — "Special Thanks Window"
Based on YouTube Super Thanks / paid comments. Supporters donate 1–2 € or more. Their message appears directly in the video — projected as fine text onto the window glass, animated with a soft glow or reflection effect. Messages rotate in a supporter queue.
Why this works:
- Creates a community-driven cozy space
- Enables continuous monetisation without ads or product placement
- Encourages repeat visits ("Is my message showing?")
- Fits the Digital Sanctuary philosophy — slow, long watch sessions
- Works well with YouTube's algorithm (high watch time = more reach)
Automation System — n8n + AI Agents
Quiz Agent — Fully Automated Pipeline
Trigger: Daily cron → reads content plan (Google Sheets / Notion)
- AI generation (LLM node): Creates question, 4 answers, correct answer, explanation, difficulty mix (20/60/20), video script, Midjourney prompt
- Image generation (Midjourney): Send prompt via Discord → wait for result → download
- Video creation: Option A — Runway or Freepik API generates the short; Option B — local FFmpeg builds the video (text overlays, light motion, optional TTS)
- Auto-publishing: n8n uploads to YouTube Shorts, Instagram Reels, Facebook Reels, TikTok → updates status + sends daily report
Lounge Radio Agent — YouTube 24/7 Cozy Stream
Trigger: Scheduled automation
- Music generation (Suno): Auto-prompt: "relaxing evening lounge" → download track → update playlist
- OBS control (WebSocket): Switch to "LoungeRadio" scene → trigger YouTube livestream start/stop
- Video scene components: Cozy interior + city view loop · music playlist player · text layer projected on window (Special Thanks)
- Special Thanks automation: n8n fetches YouTube comments / Super Thanks → formats supporter names → sends updated text to OBS via WebSocket
Outcome
A complete AI-powered content system: Strategy → Content architecture → Prompt engineering → Asset generation → Batch production → Auto-publishing.
Key takeaways:
- Two-channel strategy creates algorithmic diversity while keeping production unified
- Faceless branding enables full anonymity and long-term scalability
- The "Special Thanks Window" mechanic turns a passive watch experience into community participation
- Full n8n automation reduces weekly active work to research and quality review