Data & Analysis ✓ NVIDIA · Official defect-image-generationautomated-optical-inspectionsynthetic-datacosmos-anomalygenosmopcbausd2roi

NVIDIA Physical AI Defect Image Generation

Orchestrate defect-image generation, augmentation, inference, and labeling for AOI datasets on OSMO.

FollowSkills review · FSRS-2.0
Not recommended
55/ 100 5-point scale 2.8 / 5
Trust16 / 25 · 3.2/5

The skill requires user confirmation for flow selection, storage roots, endpoints, and pools; it documents HF_TOKEN usage and major data flows, while run names provide scoped output paths and several tasks perform failure checks. Deductions: Shell workflows download and run external content and depend on Hugging Face, GitHub, Roboflow, and cluster services; least-privilege controls, sensitive-data handling, endpoint trust validation, and general rollback are incomplete, and repository license metadata is NOASSERTION.

Reliability8 / 20 · 2.0/5

The documentation and YAML describe flows, prerequisites, inputs, outputs, parameter mappings, and many diagnosable failure checks. Deductions: key paths were not executed in this static review; reproducibility depends on container-internal scripts, OSMO, network services, external downloads, and referenced files not supplied here, with no committed key-path test suite.

Adaptability10 / 15 · 3.3/5

The skill covers PCBA, metal, glass, Day 0, Day 1, and finetuning workflows, with trigger keywords, ambiguity handling, and explicit separation of structural defects. Deductions: Chinese-language support is not declared; core operation depends on OSMO, HF, GitHub, Roboflow, and an in-cluster Qwen service without evidence of mainland-China reachability; custom scenarios still require substantial manual judgment.

Convention10 / 15 · 3.3/5

It provides tables, flow organization, prerequisites, monitoring, troubleshooting, configuration, version fields, an NVIDIA owner, and CC-BY-4.0 AND Apache-2.0 licensing. Deductions: BENCHMARK.md records missing Instructions, Examples, and email-formatted authorship; no clear changelog, maintenance commitment, or update path is shown, and many details depend on reference files not included in the supplied evidence.

Effectiveness7 / 15 · 2.3/5

The skill covers an end-to-end path from asset setup through generation, finetuning, inference, labeling, monitoring, and output retrieval, with explicit workflows, parameters, and output locations; BENCHMARK.md reports correctness and effectiveness results. Deductions: no execution was performed for this review, representative outputs cannot be independently verified, and results depend on external models, datasets, OSMO configuration, and manual prerequisites.

Verifiability4 / 10 · 2.0/5

Pinned workflow YAML, in-task validation checks, input/output paths, and an NVSkills-Eval report provide limited auditability. Deductions: the report lacks reproducible test suites, CI run artifacts, and detailed task evidence; there is no independent cross-source corroboration, and no workflow was executed here.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 55f18499943e
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Confirm dig_url_root, the OSMO pool, endpoints, and required credentials before submission; the default storage location may exceed 80 GB.
  • Core downloads and inference depend on HF, GitHub, Roboflow, OSMO, and in-cluster services; independently verify reachability, license acceptance, and dependency availability.
  • Do not upload real photos or sensitive inspection data to an unapproved URL; the supplied material does not define a general rollback or data-cleanup procedure.
  • The BENCHMARK.md results are limited evidence and do not replace key-path validation in the target environment.
Review evidence [1][2][3][4][5][6][7][8][9]
See the full review method →

What does this skill do, and when should you use it?

This skill is for teams building automated optical inspection datasets with NVIDIA Cosmos AnomalyGen. It covers PCBA Day 0 cold-start workflows and Day 1 inference and labeling, while metal-surface and glass use cases primarily follow manual-ROI workflows. OSMO workflows connect USD-to-ROI rendering, image editing, structural-defect rendering, fine-tuning, inference, and labeling. Users must prepare OSMO credentials, storage locations, and required model and dataset artifacts before submission.

Reads the selected flow, use case, checkpoint, data root, and generation parameters; submits the corresponding OSMO YAML workflow; renders PCBA USD scenes through scan-grid and ROI-crop stages; calls NVIDIA Qwen Image-Edit for appearance or lighting augmentation; uses IsaacSim for shift, tombstone, and sideflip pose defects; runs Cosmos AnomalyGen inference and inline labeling; optionally fine-tunes a checkpoint; and writes run outputs under the configured OSMO data root.

  1. An AOI team creates PCBA datasets containing texture defects, missing components, or pose defects from CAD/USD scenes.
  2. A PCBA team aligns CAD-derived rendering with a real board photo before Day 1 inference and labeling.
  3. A metal-inspection team runs inference and labeling from pre-captured clean images and ROI masks.
  4. A glass-inspection team uses the manual-ROI workflow for inference and labeling.
  5. A data engineer trains an AnomalyGen checkpoint from labeled anomaly data for later Day 0 or Day 1 use.

What are this skill's strengths and limitations?

Pros
  • Supports PCBA, metal-surface, and glass inspection scenarios.
  • Provides distinct Day 0, Day 1, and fine-tune-only workflows.
  • Documents flow selection, data handoffs, preflights, run naming, monitoring, and output locations.
  • Supports either pretrained-checkpoint passthrough or an inserted fine-tuning job.
  • The README states that published skills carry signatures, skill cards, and Tier-3 evaluation artifacts.
Limitations
  • Requires an OSMO environment for end-to-end execution; the skill file alone is not a local standalone generator.
  • Requires configured storage, credentials, and model or dataset artifacts; missing artifacts stop the workflow.
  • Day 0 requires a Qwen Image-Edit endpoint, and local deployment requires a GPU-capacity check.
  • The source does not provide independent results for this skill, operating costs, or a complete minimum hardware specification.
  • Structural-defect output scales nonlinearly with render patches, so image counts cannot be converted by simple doubling.

How do you install this skill?

Install the individual skill with the default skills CLI:

npx skills add nvidia/skills --skill physical-ai-defect-image-generation --yes

The README does not specify the exact local installation directory; the CLI prompts for or handles the destination.

How do you use this skill?

Ask an agent with the skill loaded for a specific workflow, for example: "Run the Day 0 texture-defect image-generation workflow for PCBA." Before execution, specify the flow, use case, checkpoint-versus-fine-tuning choice, and storage root; first-time use must confirm dig_url_root. Execution requires OSMO credentials, required URL artifacts, and an image-edit endpoint for Day 0. The skill requires ambiguous requests to be clarified instead of silently selecting a workflow.

FAQ

Is this suitable for generic image generation?
No. It is designed for AOI defect-image generation, augmentation, inference, and labeling, especially for PCBA, metal-surface, and glass inspection.
What must be prepared before running it?
You need OSMO credentials, a confirmed dig_url_root, accessible model and dataset URL artifacts, and the flow-specific CAD/USD, real-photo, ROI, or raw training data.
What happens if required artifacts are missing?
The skill requires a URL preflight; missing resources should be prepared through the relevant setup workflows rather than bypassed with local downloads.
Are cost and GPU requirements documented?
Pricing and complete hardware requirements are not stated. The source does specify a GPU-capacity check for local image-edit deployment in Day 0.

More skills from this repository

All from NVIDIA/skills

Data & Analysis ✓ NVIDIA · Official

PAIDF AnomalyGen

Fine-tune, generate, evaluate, and refine synthetic anomaly images.

Data & Analysis ✓ NVIDIA · Official

NV-Generate-MR

Generate synthetic body MRI volumes through NVIDIA’s rflow-mr workflow.

Data & Analysis ✓ NVIDIA · Official

NeMo Data Designer Synthetic Data Skill

Build synthetic datasets and declarative data-generation pipelines from a natural-language description.

Data & Analysis ✓ NVIDIA · Official

Data Designer Synthetic Data Skill

Build synthetic datasets and data-generation pipelines from a natural-language specification.

Automation & Ops ✓ NVIDIA · Official

OSMO Physical AI Video Augmentation Orchestrator

Safely run video augmentation, auto-labeling, and result-retrieval workflows on OSMO.

Dev & Engineering ✓ NVIDIA · Official

Clinical ASR Flywheel: Environment Setup

Validate that a clinical ASR evaluation environment can complete a TTS-to-ASR round trip through NVIDIA-hosted speech services.

Data & Analysis ✓ NVIDIA · Official

PAS People Attribute Search Workflow

Augment person-crop datasets and generate searchable attribute labels on OSMO.

Data & Analysis ✓ NVIDIA · Official

Synthetic Brain MRI Generator

Generate synthetic brain MRI volumes through NVIDIA’s documented workflow.

Automation & Ops ✓ NVIDIA · Official

Physical AI Infrastructure Resilience

Set up, validate, scale, and recover Physical AI infrastructure for synthetic data generation workflows.

Data & Analysis ✓ NVIDIA · Official

TAO AOI Image Mining

Embed target and source images with one encoder, then mine deduplicated nearest-neighbour AOI images for augmentation.

Data & Analysis ✓ NVIDIA · Official

TAO Optical Inspection Training

Train and deploy Siamese-network models for manufacturing defect inspection.

Data & Analysis ✓ NVIDIA · Official

TAO Visual ChangeNet Inspection

Train paired-image classifiers and change-segmentation models for PCB defect inspection.

Data & Analysis ✓ NVIDIA · Official

TAO ChangeNet Root Cause Analyzer

Uses image evidence to identify why NVIDIA TAO Visual ChangeNet classification experiments fail.

Data & Analysis ✓ NVIDIA · Official

NV-Generate-CT Synthetic CT Generator

Generate paired synthetic CT volumes and masks through NVIDIA’s rflow-ct pipeline.

Data & Analysis ✓ NVIDIA · Official

TAO DEFT AOI Improvement Loop

Automates evaluation, defect augmentation, mining, retraining, and deployment gating for PCB inspection models.

Dev & Engineering ✓ NVIDIA · Official

Jetson BSP Info Check

Verify Jetson BSP version, board configs, and rootfs state before flashing.

Dev & Engineering ✓ NVIDIA · Official

DALI Dynamic Mode Assistant

Helps agents write, review, and migrate NVIDIA DALI imperative dynamic-mode code.

Automation & Ops ✓ NVIDIA · Official

VSS Video Embedding Deployment

Deploy and operate NVIDIA’s video embedding service for files, text, and live streams.

Automation & Ops ✓ NVIDIA · Official

DOCA Telemetry Utils

Discover telemetry counters, translate Data IDs, and verify BlueField support before committing exporter configuration.

Data & Analysis ✓ NVIDIA · Official

Earth2Studio Weather Data Fetch

Fetch validated weather and climate variables from Earth2Studio sources by time and data type.

Related skills