Uploaded June 2026 | Updated September 2026, 2 weeks ago
In this tutorial, you will learn how to set up and use NVIDIA physical AI agent tools and skills for defect image generation to create synthetic anomaly images for visual inspection pipelines. The Defect Image Generation skill runs in an NVIDIA Brev launchable that helps reduce months of manual defect data collection and labeling — you describe the defect you want in natural language and the agent generates labeled anomaly images ready to train your inspection model. By the end of the tutorial, you will have the Defect Image Generation skill running and be able to generate synthetic defect datasets for PCB, glass, metal, and your own custom use cases.
Get started with the Defect Image Generation skill:
🚀 Deploy the NVIDIA Brev Launchable → brev.nvidia.com/launchable/deploy?launchableID=env-3DMiyRSFam5te8UXrCniX9QCu3x
⬇️ Download the open-source skill on GitHub → github.com/NVIDIA/skills/tree/main/skills/physical-ai-defect-image-generation
Chapters:
00:00 Intro — Navigate to brev.nvidia.com/physical-ai
00:15 Defect Image Generation Skill — architecture overview
00:35 Select and configure the Defect Image Generation launchable
01:05 Set up credentials (NVIDIA API key + Hugging Face token)
01:35 Open OpenClaw and send setup prompt
02:00 Agent reports blocked credentials — provide NGC and HF keys
02:20 New session — send PCBA bridge defect generation prompt
02:40 Agent submits workflow to OSMO and reports pipeline status
03:00 Ask agent for status update
03:10 View output — input image, mask, and reconstructed defect
#NVIDIAAgentSkills #PhysicalAI #SyntheticData
NVIDIA Agent Skills, synthetic data generation, defect detection, visual inspection, Physical AI, NVIDIA Launchable
Q: What is the NVIDIA Defect Image Generation skill?
A: It is a launchable workflow that uses a synthetic data generation agent to create labeled anomaly images for visual inspection models. You describe the defect you want in natural language and the agent runs the required scripts and models automatically, turning months of data collection into a few hours of compute.
Q: What defect types are supported out of the box?
A: The launchable includes pre-trained models and data for PCB, glass, and metal defects. You can also extend it to custom defect types for your own inspection model.
Q: What API key do I need to get started?
A: You can use an API key from build.nvidia.com, OpenAI, or Anthropic. You will also need a Hugging Face API key and an NGC API key so the agent can pull the required models and datasets.
Q: How long does it take to generate defect images?
A: Generation time depends on the number of images and your GPU configuration. In this walkthrough, a single defect image generated in approximately 10 minutes. Large-scale batch generation is also supported.
Q: What is OSMO and why does it appear in the workflow?
A: OSMO is the backend orchestration tool the agent uses to manage and run tasks. It shows a history of all your generation requests, real-time status updates, and logs useful for debugging — you can check it any time during a generation run.
In this tutorial, you will learn how to set up and use NVIDIA physical AI agent tools and skills for defect image generation to create synthetic anomaly images for visual inspection pipelines. The Defect Image Generation skill runs in an NVIDIA Brev launchable that helps reduce months of manual defect data collection and labeling — you describe the defect you want in natural language and the agent generates labeled anomaly images ready to train your inspection model. By the end of the tutorial, you will have the Defect Image Generation skill running and be able to generate synthetic defect datasets for PCB, glass, metal, and your own custom use cases.
Get started with the Defect Image Generation skill:
🚀 Deploy the NVIDIA Brev Launchable → brev.nvidia.com/launchable/deploy?launchableID=env-3DMiyRSFam5te8UXrCniX9QCu3x
⬇️ Download the open-source skill on GitHub → github.com/NVIDIA/skills/tree/main/skills/physical-ai-defect-image-generation
Chapters:
00:00 Intro — Navigate to brev.nvidia.com/physical-ai
00:15 Defect Image Generation Skill — architecture overview
00:35 Select and configure the Defect Image Generation launchable
01:05 Set up credentials (NVIDIA API key + Hugging Face token)
01:35 Open OpenClaw and send setup prompt
02:00 Agent reports blocked credentials — provide NGC and HF keys
02:20 New session — send PCBA bridge defect generation prompt
02:40 Agent submits workflow to OSMO and reports pipeline status
03:00 Ask agent for status update
03:10 View output — input image, mask, and reconstructed defect
#NVIDIAAgentSkills #PhysicalAI #SyntheticData
NVIDIA Agent Skills, synthetic data generation, defect detection, visual inspection, Physical AI, NVIDIA Launchable
Q: What is the NVIDIA Defect Image Generation skill?
A: It is a launchable workflow that uses a synthetic data generation agent to create labeled anomaly images for visual inspection models. You describe the defect you want in natural language and the agent runs the required scripts and models automatically, turning months of data collection into a few hours of compute.
Q: What defect types are supported out of the box?
A: The launchable includes pre-trained models and data for PCB, glass, and metal defects. You can also extend it to custom defect types for your own inspection model.
Q: What API key do I need to get started?
A: You can use an API key from build.nvidia.com, OpenAI, or Anthropic. You will also need a Hugging Face API key and an NGC API key so the agent can pull the required models and datasets.
Q: How long does it take to generate defect images?
A: Generation time depends on the number of images and your GPU configuration. In this walkthrough, a single defect image generated in approximately 10 minutes. Large-scale batch generation is also supported.
Q: What is OSMO and why does it appear in the workflow?
A: OSMO is the backend orchestration tool the agent uses to manage and run tasks. It shows a history of all your generation requests, real-time status updates, and logs useful for debugging — you can check it any time during a generation run.
