Open-source local image-to-3D pipeline keeps text and logos after retopo, 900k faces down to ~5k
Bingeljell · reddit · 2026-10-01
Reddit user Bingeljell open-sourced a fully local image-to-3D pipeline aiming at Tripo/Meshy-quality output without cloud APIs.
- New Pixel Match feature: copies real pixels from the source image back onto visible model surfaces, so text, logos and fine details like "VANGUARD 07" survive retopology from 900k faces down to 5k
- Workflow: prompt → Qwen-Image 2.1 → Pixal3D → Finish (Pixel Match) → textured GLB; produces game-ready and sprite-ready assets
- Backends: Pixal3D (8.6GB, Mac + NVIDIA), Stable Fast 3D (fast, lower detail, gated weights), with TRELLIS.2 and Hunyuan3D running on Mac
- Platforms: NVIDIA (Linux tested, Windows limited) and Apple Silicon, one-line installer, Blender 4.2+ needed for Finish
The author built it to automate asset creation for their own game; rigging/animation are still WIP. Repo and v0.3.5 release notes on GitHub.
More from Multimodal
- Seedance 2.5 workflow: preview 5 takes at 480p, render only the winner at 1080p — alifcoder · 2026-10-01
- Midjourney announces its weekly Office Hours livestream for September 30 — midjourney · 2026-10-01
- techhalla shares structure-preserving photoreal workflow with GPT 2.5 Sunburst — techhalla · 2026-10-01
- techhalla turns a 2D floor plan into pro real estate assets with Magnific + Opus 5.5 — techhalla · 2026-10-01
- 3D artist calls GPT Astra the best 3D modeler he's met after dense tunnel scene — TomLikesRobots · 2026-10-01
- AI short film 'Inside STILL' moves from Bryant Park in rain to NYC Library's Rose Reading Room — Daniel_Farinax · 2026-10-01