25 lines
1.1 KiB
Markdown
25 lines
1.1 KiB
Markdown
# Ollama Vision Models for Home Assistant
|
|
|
|
Use these models for camera snapshot analysis (e.g., Motion Front Door AI automation).
|
|
|
|
| Model | Size | Command | Best For |
|
|
|-------|------|---------|----------|
|
|
| llama3.2-vision:11b | ~7.8GB | `ollama pull llama3.2-vision:11b` | Scene/object description, general vision |
|
|
| moondream | ~1.7GB | `ollama pull moondream` | Tiny and fast, simple descriptions |
|
|
| gemma4:12b | ~8GB | `ollama pull gemma4:12b` | Newest, best quality, all-rounder |
|
|
| llava:13b | ~8GB | `ollama pull llava:13b` | Older but proven, general-purpose |
|
|
|
|
## How to switch models
|
|
|
|
1. Pull the model: `ollama pull <model-name>`
|
|
2. Go to **Settings → AI Tasks → Ollama AI Task** in Home Assistant
|
|
3. Change the model to the new one
|
|
4. Test by triggering the automation
|
|
|
|
## Notes
|
|
|
|
- Vision models are required for image/camera analysis - text-only models (llama3, qwen3, etc.) will fail
|
|
- `llama3.2-vision:11b` is the recommended default (best balance of quality and speed)
|
|
- `moondream` is a good fallback for slower hardware
|
|
- All models run locally - no data leaves your network
|