[Custom Integration] HA Video Vision - AI Video Analysis + Facial Recognition

Share your Projects! Custom Integrations
,
1

HA Video Vision

AI-powered video analysis and facial recognition for Home Assistant cameras.

Installation

hacs_badge

Add as custom repository in HACS:
https://github.com/LosCV29/ha-video-vision

Features

  • :movie_camera: Real video analysis - sends video clips, not snapshots
  • :bust_in_silhouette: Facial recognition - “Who’s at the door?” → “It’s Carlos”
  • :free: Free by default - OpenRouter Nemotron model
  • :house: Local option - run on your own GPU
  • :iphone: Smart notifications - AI descriptions with snapshots

Supported Providers

Provider Model Cost
OpenRouter Nemotron 12B VL FREE
Google Gemini gemini-2.0-flash Free tier
Local vLLM Qwen-VL, LLaVA Free

Example Output

Instead of: “Motion detected on Front Porch”

You get: “A woman in a blue jacket is approaching the front door carrying a package. Identified: Mom (87% confidence)”

Works With

  • PolyVoice - Voice control: “Check the driveway”
  • Any RTSP camera - Reolink, Hikvision, etc.
  • Frigate - Trigger on events

Links

Feedback welcome!

1 Like
2

Looks super promising, looking forward to check it out. Am I correct in understanding that for those of us without a dedicated machine with a graphics card, this can still be used to send alerts if the camera spots say “a man carrying a parcel”? Just not do actual face recognition? Would an Nvidia Jetson Nano be useful for something like this - or is something beefier required?

Any chance of supporting Google Coral in the future like frigate does? Or does that not help here?

3

My apologies… As this is LLM based, even small models would be too much for a coral or Nano. If you want to try the features, you can definitely try with Open Routers free models. They work great - ive tested it.

4

Ah I see. So it’s not even possible to point the facial recognition at Gemini, for instance, to do the work?

5

The facial rec is a separate component. But its very light. The facial rec can be done by a CPU, no beefy equipment needed. Just spin up a server on any PC and point the integration at your facial rec server. You can run a mix of cloud for video capture and analysis and local for facial rec component. Hope that helps.

1 Like