Honeycomb: constant-size HexMemory keeps video world models consistent over long-horizon generation
Jack Wei Lun Shi · hf · 2026-10-03
Honeycomb is a video world model built on HexMemory, a low-rank scene representation stored in six fixed-size spatial and spatiotemporal planes. A feed-forward writer maps each generated chunk into plane features; as coverage expands, previous planes are warped dimension-preserving and fused via confidence-weighted pooling and learned residual correction. A reader retrieves latents to condition generation, processing only new chunks—no per-scene optimization or full-history reprocessing. Experiments on WorldScore and RealEstate10K show strong quality and revisit consistency with constant memory. Code is open.
More from Multimodal
- Runway's Praxis-1 robot brain learns from ordinary videos, cutting demo data from thousands of hours to minutes — c_valenzuelab · 2026-10-03
- Bona Film's 100-minute AI-assisted feature 'Sanxingdui: Future Past' gets theatrical release Oct 23 — lmoroney · 2026-10-03
- AI Director Junie Lau Featured in Forbes; Goldfish Mayhem Premieres with Wonder Studios — JunieLauX · 2026-10-03
- fal's H3 Max Reference-to-Video adds first, middle, and last frame control — OdinLovis · 2026-10-03
- Developer Runs Blender Inside ChatGPT Dots to Auto-Model, Rig and Animate a Dance Video — OpenAIDevs · 2026-10-03
- Patching ComfyUI-MPS-INT8 to fix MiniMax H3 VAE crashes on Mac — Kris_RenderBob · 2026-10-03