Blur2Vid: Generating Videos from Motion-Blurred Images
cindy_x_wu · x · 2026-08-10
Researchers from University of Toronto and York University introduced Blur2Vid, accepted to SIGGRAPH Asia 2025.
- Core Capability: It generates videos from single motion-blurred images, uniquely predicting the past, present, and future frames within the exposure time.
- Applications: The model excels at reconstructing complex in-the-wild motions and bringing historical photos to life.
- Technical Extensions: When combined with MegaSAM, the generated videos can be used to recover dynamic 3D reconstructions and 3D human poses.
More from Multimodal
- Developer Creates Stunning Interactive 3D Alpine Valley Using Three.js — techartist_ · 2026-08-10
- RTX 5080 vs. 4090: Which is Better for Running Minimax Locally? — ASK_ABT_MY_USERNAME · 2026-08-10
- Maximizing Identity Preservation in MiniMax H3 img2vid — NetworkSpecial3268 · 2026-08-10
- Help Needed: LTX 2.3 Long Video Generation is Too Slow (750s/clip) — zer0mz · 2026-08-10
- Seeking a Stable Workflow for Reference-to-Video Character Swap — Wonderful-Tough7215 · 2026-08-10
- Testing Seedance 2.5: Generating 30s Videos with Multiple References — CurieuxExplorer · 2026-08-10