{"id":773,"date":"2025-07-31T05:50:37","date_gmt":"2025-07-31T05:50:37","guid":{"rendered":"https:\/\/blog.aetherix.com\/?p=773"},"modified":"2026-03-27T13:18:30","modified_gmt":"2026-03-27T13:18:30","slug":"jetson-generative-ai-live-llava","status":"publish","type":"post","link":"https:\/\/blog.openzeka.com\/en\/jetson-generative-ai-live-llava\/","title":{"rendered":"Jetson Generative AI \u2013 Live LLaVA"},"content":{"rendered":"
\n
Vision-Language Models reach new heights when applied to live video streams<\/strong>\u2014Live LLaVA<\/strong> demonstrates real-time multimodal AI that can see, understand, and describe what’s happening in your camera feed continuously<\/strong>\u00a0on your Jetson device.<\/div>\n
<\/div>\n
In this article you’ll learn how to run Live LLaVA with optimized vision-language models like LLaVA and VILA, featuring hardware-accelerated video processing and real-time inference capabilities.<\/div>\n<\/div>\n

 <\/p>\n<\/div>

Requirements<\/h3><\/div>\n
\n

 <\/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n
\n
\n
Hardware \/ Software<\/div>\n<\/div>\n<\/th>\n
\n
\n
Notes<\/div>\n<\/div>\n<\/th>\n<\/tr>\n<\/thead>\n
\n
\n
Jetson AGX Orin (64GB)<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Recommended for best performance<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
Jetson AGX Orin (32GB)<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Good performance for most use cases<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
Jetson Orin NX (16GB)<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Solid performance<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
Jetson Orin Nano (8GB)<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Minimum requirement – use smaller models<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
JetPack 6 (L4T r36.x)<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Required for latest optimizations<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
USB camera or CSI camera<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
For live video input<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
NVMe SSD highly recommended<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
For storage speed and space<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
22GB for nano_llm container<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Container image storage<\/div>\n<\/div>\n<\/td>\n<\/tr>\n
\n
\n
>10GB for models<\/strong><\/div>\n<\/div>\n<\/td>\n
\n
\n
Vision-language model storage<\/div>\n<\/div>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n
\n
Note:<\/strong>\u00a0Follow the NanoVLM tutorial first to familiarize yourself with vision\/language models, and see Agent Studio for an interactive pipeline editor.<\/em><\/div>\n<\/div>\n<\/div>

\n

Supported Models<\/h4>\n<\/div><\/h1><\/div>

The following vision-language models are optimized for Live LLaVA:<\/p>\n

\n
\n
LLaVA Models:<\/strong><\/div>\n
\n
`liuhaotian\/llava-v1.5-7b`,<\/div>\n
`liuhaotian\/llava-v1.5-13b`<\/div>\n
`liuhaotian\/llava-v1.6-vicuna-7b`<\/div>\n
`liuhaotian\/llava-v1.6-vicuna-13b`<\/div>\n<\/blockquote>\n
VILA Models:<\/strong><\/div>\n
\n
`Efficient-Large-Model\/VILA-2.7b`<\/div>\n
`Efficient-Large-Model\/VILA-7b`<\/div>\n
`Efficient-Large-Model\/VILA-13b`<\/div>\n
`Efficient-Large-Model\/VILA1.5-3b`<\/div>\n
`Efficient-Large-Model\/Llama-3-VILA1.5-8B`<\/div>\n
`Efficient-Large-Model\/VILA1.5-13b`<\/div>\n<\/blockquote>\n
Jetson Orin Nano Compatible Models:<\/strong><\/div>\n
\n
VILA-2.7b<\/div>\n
VILA1.5-3b<\/div>\n
VILA-7b<\/div>\n
Llava-7b<\/div>\n
Obsidian-3B<\/div>\n<\/blockquote>\n<\/div>\n<\/div>\n<\/div>

\n

Step-by-Step Setup<\/h3>\n<\/div><\/h3><\/div>

\n

1. Verify Camera Connection<\/h4>\n<\/div><\/h1><\/div>

Check that your camera is properly connected and detected:<\/p>\n<\/div>