JoVox: image-conditioned hierarchical flow matching generates 64³ voxels in one model
pmttyji · reddit · 2026-11-10
JoVox (Joint Hierarchical Flow Matching for Lightweight Voxel Generation) is now on Hugging Face: an image-conditioned 64³ voxel structure generator.
It uses a discrete flow matching framework that learns the joint hierarchical distribution of voxels while staying lightweight — a single model generates all levels, 4³ → 16³ → 64³, from one image.
Paper (arXiv 2611.10690), code, and project page with benchmarks are all publicly available.
More from Multimodal
- MiniMax teases upcoming H3-Image generation model — victormustar · 2026-11-10
- Dev builds vocal synth modeling the vocal tract as branching acoustic tubes, not phoneme stitching — TinfoilTricorn · 2026-11-10
- Open-source RGBMocap adds live mocap streaming into Blender with motion baking — camenduru · 2026-11-10
- Local MiniMax H3 on 6GB VRAM recreates Harry Potter scene at ~50 min per image — No_Cow26 · 2026-11-10
- 3D celebrity caricature prompt shared: exaggerated head, recognizable face — azed_ai · 2026-11-10
- Agent given a vocal-synth repo builds branching acoustic-tube singing synthesis — TinfoilTricorn · 2026-11-10