# physical-ai-video-data-augmentation

Runs the video augmentation and auto-labeling loop on OSMO, from picking a flow and preflight checks to monitoring and pulling outputs.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 2000 installs · ★ 3735
- Author: NVIDIA
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/NVIDIA/skills/tree/HEAD/skills/physical-ai-video-data-augmentation/SKILL.md
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/skills/physical-ai-video-data-augmentation

## What it does

This skill handles OSMO VDA runs so you do not have to chain the steps by hand. It maps your request to one of four flows: `auto_labeling`, `augmentation_and_al`, `e2e`, `e2e_super_resolution`. Then the agent walks the run from preflight to downloaded results.

Several scripts do the work. Credentials and image access are checked by `preflight_credentials.sh`. Interpolation values, cache state and dataset safety pass through `pre_submit_guard.py`. Demo clips come from `prepare_demo_assets.sh`. Separate workers cover augmentation and labeling, a barrier script keeps multiple nodes in step, and a renderer builds a side-by-side view of augmented flows.

Three things must be in place. The `osmo` CLI has to be on `PATH`, logged in, with a DATA credential profile. A Hugging Face token is needed for the gated Cosmos and SeedVR weights. And `osmo pool list --mode free` must show at least one ONLINE GPU pool, otherwise nothing gets scheduled. An NGC API key is optional. The skill also tells the agent to read `storage_url` from the active dataset backend instead of guessing it; a wrong scheme breaks storage auth at runtime.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill physical-ai-video-data-augmentation -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/physical-ai-video-data-augmentation/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/physical-ai-video-data-augmentation/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill physical-ai-video-data-augmentation -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/physical-ai-video-data-augmentation/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill physical-ai-video-data-augmentation -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/physical-ai-video-data-augmentation/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install NVIDIA/skills physical-ai-video-data-augmentation
```

2. **Manually: put the files in place**
- File: `.github/skills/physical-ai-video-data-augmentation/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill physical-ai-video-data-augmentation -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/NVIDIA/skills --path skills/physical-ai-video-data-augmentation
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/physical-ai-video-data-augmentation/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill physical-ai-video-data-augmentation -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/NVIDIA/skills/tree/HEAD/skills/physical-ai-video-data-augmentation/SKILL.md

## When to use it
- match a VDA request to the right flow
- verify secrets and image access before submitting
- take `storage_url` from the dataset backend, not from guesswork
- follow a workflow through to finished outputs
- render a before-and-after view of augmented video

## Example prompts
- Label my source videos only, use the `auto_labeling` flow
- Start the full VDA pipeline on OSMO and list the GPU pools you can target
- Augment the videos, auto-label the outputs, then render a side-by-side comparison

## Similar
- [skill-creator](https://affpapa.org/en/nevibecode/skills/skill-creator.md) — skill · 414180 installs
- [writing-for-agents](https://affpapa.org/en/nevibecode/skills/writing-for-agents.md) — skill · 466719 installs
- [media-use](https://affpapa.org/en/nevibecode/skills/media-use.md) — skill · 1100000 installs
- [mcp-builder](https://affpapa.org/en/nevibecode/skills/mcp-builder.md) — skill · 130949 installs
- [subagent-driven-development](https://affpapa.org/en/nevibecode/skills/subagent-driven-development.md) — skill · 237746 installs
- [orchestration](https://affpapa.org/en/nevibecode/skills/orchestration.md) — skill · 349800 installs

From the same set: [nvidia/skills](https://affpapa.org/en/nevibecode/by/nvidia/skills.md) — 195 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md