Image Generation: - Three-tier visual generation: Illustrate (ComfyUI ~8s), 3D Render (Blender ~20s), AI→3D Sculpt (TripoSR ~60s) - TripoSR image-to-3D mesh pipeline: ComfyUI generates concept → TripoSR creates 16K+ vertex mesh → Blender renders gold sculpture - Blender 4.3.2 with EEVEE renderer, 12 procedural scene types (nautilus, clockwork, neural mesh, fractal tree, Möbius, cathedral, tesseract, wave field, lattice cage, bridge, galaxy, toroidal knot) - Golden ratio composition: 1280x320 panoramic banners, fibonacci spiral prompts - 50 steps, cfg 8.5, DPM++ 2M Karras for ComfyUI quality - VRAM management: auto-stops ComfyUI for TripoSR, restarts after - Displacement map fallback if TripoSR unavailable - All images cached to disk — instant on repeat UI Fixes: - Mobile responsive: wrapped nav, compact cards, tight spacing - Mermaid.js fully removed (caused visible errors) - One-shot illustrate buttons (no stacking) - Button labels: Illustrate / 3D Render / AI→3D Sculpt - Loading states show pipeline steps Infrastructure: - UFW rules for LAN access on :3500, :3600, :8188 - Boot order: PostgreSQL+Ollama → LLM Team → ComfyUI → imagegen - Demo/showcase mode auto-starts on boot - :5000 stays 127.0.0.1 (nginx proxied) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Description
LLM Team UI - Full-stack local AI orchestration platform
Languages
Python
97.4%
Shell
2.6%