Dev trains a diffusion model on optical flow maps from video
pixlpa · x · 2026-09-11
pixlpa shares an experiment: extracting optical flow maps from video clips and training a diffusion model on them — a small hands-on exploration of diffusion models learning motion representations, with training samples attached.
More from Multimodal
- pixel2motion: open-source skill turns raster logos into smooth SVG animations (2.2k stars) — tom_doerr · 2026-09-11
- invideo partners with Sony Pictures to stream AI-made originals on SonyLIV — gen_ericai · 2026-09-11
- Unsiloed turns time-series charts into structured data agents can compute with — bigaiguy · 2026-09-11
- GPT-6 Astra demo: generating 3D objects, key-framing and lighting VFX in After Effects — jxnlco · 2026-09-11
- "AI isn't making us lazy" sparks debate — with the full Midjourney recipe — umesh_ai · 2026-09-11
- ChatGPT Prompt Puts You in Any Location With Matched Lighting — ifioknkem · 2026-09-11