Uploaded October 2026 | Updated November 2026, 1 day ago
π οΈ Build computer vision workflows without code β pysource.com
Nvidia Cosmos on github: github.com/nvidia/cosmos
The demo I used in this video: github.com/pysource-com/Cosmos-Edge-Studio
In this video, I test NVIDIA Cosmos on the new NVIDIA Jetson AGX Thor to see how far AI can go beyond traditional computer vision.
Instead of simply detecting objects, counting vehicles, or tracking movement, NVIDIA Cosmos is designed to understand what is happening inside a scene, reason about it, and even predict what might happen next.
I test Cosmos Reasoner on different real-world traffic videos and ask questions such as:
β’ What is happening in this scene?
β’ Are there any traffic violations?
β’ Is there anything unusual?
β’ What might happen next?
β’ Can the AI identify potential risks before they happen?
For example, we test whether the model can understand wet road conditions, reason about vehicles and pedestrians, interpret unusual traffic situations, and make predictions based on the visual context.
Everything runs locally on the NVIDIA Jetson AGX Thor, which features 128 GB of memory and is designed for running large AI models directly at the edge.
I also built a simple interface called Cosmos Edge Studio to make it easier to load videos, interact with NVIDIA Cosmos models, and experiment with video reasoning. I plan to make the project available on GitHub.
This is only the beginning of my testing. In future videos, I can compare the different Cosmos models, including Edge, Nano, and Super, and push the system with more difficult CCTV and traffic scenarios.
Let me know in the comments what videos or situations you would like me to test next.
π Learn more about PySource β pysource.com
#ComputerVision #NVIDIA #JetsonThor #NVIDIACosmos #PhysicalAI #ArtificialIntelligence #EdgeAI #VideoAnalytics #TrafficAnalysis
π οΈ Build computer vision workflows without code β pysource.com
Nvidia Cosmos on github: github.com/nvidia/cosmos
The demo I used in this video: github.com/pysource-com/Cosmos-Edge-Studio
In this video, I test NVIDIA Cosmos on the new NVIDIA Jetson AGX Thor to see how far AI can go beyond traditional computer vision.
Instead of simply detecting objects, counting vehicles, or tracking movement, NVIDIA Cosmos is designed to understand what is happening inside a scene, reason about it, and even predict what might happen next.
I test Cosmos Reasoner on different real-world traffic videos and ask questions such as:
β’ What is happening in this scene?
β’ Are there any traffic violations?
β’ Is there anything unusual?
β’ What might happen next?
β’ Can the AI identify potential risks before they happen?
For example, we test whether the model can understand wet road conditions, reason about vehicles and pedestrians, interpret unusual traffic situations, and make predictions based on the visual context.
Everything runs locally on the NVIDIA Jetson AGX Thor, which features 128 GB of memory and is designed for running large AI models directly at the edge.
I also built a simple interface called Cosmos Edge Studio to make it easier to load videos, interact with NVIDIA Cosmos models, and experiment with video reasoning. I plan to make the project available on GitHub.
This is only the beginning of my testing. In future videos, I can compare the different Cosmos models, including Edge, Nano, and Super, and push the system with more difficult CCTV and traffic scenarios.
Let me know in the comments what videos or situations you would like me to test next.
π Learn more about PySource β pysource.com
#ComputerVision #NVIDIA #JetsonThor #NVIDIACosmos #PhysicalAI #ArtificialIntelligence #EdgeAI #VideoAnalytics #TrafficAnalysis










