A slimmed MiniMax-H3 that makes video with sound on a single 24 GB card
Emerged finds new products worth a look, daily — what’s new today
tamserve
Serve SAM 2-family tracking for many live cameras from one model on one GPU
Is this yours? Claim itIf you track objects across many camera feeds, sharing one loaded model can cut GPU memory several-fold and the hardware bill with it.
tamserve is a Python engine for real-time video object tracking that loads a single EdgeTAM, SAM 2.1 or optional SAM 3 model and multiplexes every camera through it. On one RTX 4090 the author reports 8 cameras at 323 frames/s in 2.2 GB versus 8.6 GB for the reference code, with matching DAVIS accuracy, and 64 cameras in 5.7 GB. SAM 3 extras add tracking by text name or people. Install with pip; Apache-2.0, with demo clips and Colab notebooks.
More like this
One tap on your phone gives you a vertical and a wide video, both full frame
Lap, sector and corner analysis for track-day GPS logs, with video overlays
Dollar-per-second prices for AI video models across APIs and credit plans
Webcam person detection where the cloud server never sees a single frame
A tuned vLLM stack that serves Qwen3.8-27B fast on one RTX 3090