README.md

metadata

title: Immersive Vibe Development Studio
emoji: 🎮
colorFrom: purple
colorTo: green
sdk: gradio
sdk_version: 6.13.0
app_file: app.py
hf_oauth: true
hf_oauth_scopes:
  - inference-api
pinned: true
license: mit

🎮 Immersive Vibe Development Studio

Pick 2–5 frontier AI models, type a 2-word theme (e.g. jungle monkey), and watch them collaboratively build a playable Three.js platformer — live, in your browser.

While the game is being generated, an isometric Game Dev Story-style studio plays out: each model gets a character at a desk, speech bubbles show their output in real time, and the scene reacts to each pipeline phase. When done, the result monitor zooms in and the game runs directly in the canvas.

---

How it works

1. Log in with your HuggingFace account (OAuth — your token is used only for Inference API calls, never stored)
2. Type a 2-word trigger — theme hero, e.g. space robot, medieval knight, ocean dolphin
3. Select 2–5 models — roles are assigned automatically (fastest → Art Director, most powerful → Lead Coder)
4. Click Generate and watch the studio come alive (~15–30 min)
5. Download the self-contained ZIP when done — open index.html locally, no server needed

---

Pipeline phases

Phase	Role	What happens
1	Art Director	Asset manifest — 9 assets, palette, silhouettes
2	Char Designer	Hero geometry spec (box primitives)
3	Geom Builder	Three.js geometry for all assets
4	Texture Director	Canvas2D procedural textures
5	Game Architect	Physics, mechanics, tile recycling
6	Scene Composer	Lighting, fog, atmosphere
7	Lead Coder	Complete main.js initial build
8	Lead Coder + Geom Builder	Round-robin refinement (4 rounds)
9	QA	Auto-lint + one auto-fix pass

---

Download contents

File	Description
index.html	The complete playable game (file://-safe, CDN Three.js)
trace.html	Full pipeline trace with per-phase timing

---

Local development

git clone https://huggingface.co/spaces/BladeSzaSza/Immersive-Vibe-Development-Studio
cd Immersive-Vibe-Development-Studio

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"          # or: pip install -r requirements.txt

# Re-extract character pool from your own playground games (optional):
python scripts/extract_characters.py

# Run locally (OAuth mocked — HF CLI login required):
huggingface-cli login
python app.py
# → open http://localhost:7860

Note: OAuth features are mocked outside HF Spaces. Run huggingface-cli login first so local mode can resolve your token.

---

Tech stack

• Gradio 6 — UI, OAuth, streaming
• Three.js r167 — isometric studio scene + generated game (CDN)
• HuggingFace Inference API — all model calls (no GPU required)
• openfloor — protocol types for trace formatting