DiffusionGemma's native vision tower runs near-real-time object detection on a phone
bodonoghue85 · x · 2026-09-19
Developer mmastrac demonstrated "djev" (DiffusionGemma-as-Jev): running near-real-time vision object detection on a mobile phone using DiffusionGemma's native vision tower.
The clip shows the small model bounding objects in real time, highlighting how its built-in vision capability can power on-device detection without extra tooling.
More from Fun
- Running 288k context at 5 tokens/sec for 16 hours: "Maybe I am Qwen" — LeviTurk · 2026-09-19
- Meme: What Big Tech engineering org charts look like in 2026 — venturetwins · 2026-09-19
- Frontier agents play tag in a self-built NYC sim as Fable 5.1 beats Astra 2-1 — Lianhuiq · 2026-09-19
- "7 out of 10 Jev projects on X are useless": pushback against hype demos — RexDouglass · 2026-09-19
- Chile's government published a map of towns in wrong places, likely AI-generated — PolarBearby · 2026-09-19
- Bragging about burning $30k in tokens for one AI video is the new social media flex — AIandDesign · 2026-09-19