# Tao finetune video clip

> Tao finetune video clip is an open-source agent skill for design and media. Runs InternVideo2-CLIP video-text retrieval, fine-tuning, ONNX export and TensorRT deploy. It is available on Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf and Goose.

- **Kind:** Skill
- **Pricing:** Open source
- **Platforms:** Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf, Goose
- **Website:** https://github.com/NVIDIA/skills/tree/main/skills/tao-finetune-video-clip
- **Repository:** https://github.com/nvidia/skills
- **GitHub stars:** 4,037
- **License:** Apache-2.0
- **Skill name:** tao-finetune-video-clip
- **Collection:** https://github.com/NVIDIA/skills
- **Categories:** design and media, AI models and LLM tooling
- **Listing updated:** 2026-11-18

## About

This skill guides an agent through the NVIDIA TAO video_clip workflow built on the InternVideo2-CLIP L14 model. It covers training, evaluation, embedding inference and ONNX export in the TAO PyTorch container, with LoRA options for vision-side fine-tuning. A separate TAO Deploy path handles TensorRT engine generation and TensorRT-backed evaluation and inference after export.

The skill sets out a preflight sequence that checks the container image is complete, including the video_clip import and PyAV availability, before any data is pulled or a run is launched. It also lays out a expected workspace layout with metadata JSON in the vadr1_chunks format, spec templates, and bind-mounted data, model and results folders. Offline evaluation expects MobileCLIP and InternVideo2 weights to be staged on disk, and Hugging Face tokens are only needed for online downloads.

The agent needs Docker with nvidia-container-toolkit and an NVIDIA GPU, or a local tao-pytorch checkout with a tao-cli virtualenv. Video records must use absolute video paths, and the skill includes smoke-test overrides and notes on common pitfalls such as precision settings and sidecar files required for deployment.

## When your agent uses it

> InternVideo2-CLIP L14 (TAO video_clip) for video-text retrieval, zero-shot classification, embedding extraction, LoRA fine-tuning, ONNX export, and TensorRT deployment. Use when the user asks to "fine-tune IV2CLIP", "run video_clip train/evaluate/inference/export", "build a Video-CLIP TensorRT engine", "InternVideo2-CLIP on KPI chunks", or "TAO video_clip on vadr1_chunks JSON".

## What it can access

From a static check of the files on GitHub; nothing was run.

- [Instructions only, no scripts](https://github.com/NVIDIA/skills/tree/HEAD/skills/tao-finetune-video-clip)
- [Asks for: Read, Bash](https://github.com/NVIDIA/skills/blob/HEAD/skills/tao-finetune-video-clip/SKILL.md)
- [License: Apache-2.0](https://github.com/NVIDIA/skills/tree/HEAD/skills/tao-finetune-video-clip)

## Features

- **Training and fine-tuning:** Runs video_clip train with spec overrides, including LoRA configuration for vision-side adaptation.
- **Retrieval and classification evaluation:** Scores video-text retrieval or zero-shot classification on a validation split.
- **Embedding inference:** Writes video and text embeddings to HDF5 files from a gallery of clips and text prompts.
- **ONNX export:** Exports a combined image-and-text ONNX model along with config and tokenizer sidecars.
- **TensorRT deployment:** Builds TensorRT engines and runs evaluation and inference through the separate TAO Deploy image.
- **Import and environment checks:** Verifies the container image is complete and that PyAV is present before a run starts.

## Good for

- Fine-tune InternVideo2-CLIP on a set of labeled video clips
- Evaluate video-text retrieval accuracy on a validation split
- Generate video and text embeddings for a clip gallery
- Export a trained Video-CLIP checkpoint to ONNX
- Build a TensorRT engine from the exported model

## Install

### Claude Code

Copy the skill folder to `~/.claude/skills/tao-finetune-video-clip` to use it in all your projects:

```sh
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.claude/skills &&
rm -rf ~/.claude/skills/tao-finetune-video-clip &&
cp -R "$tmp/skills/tao-finetune-video-clip" ~/.claude/skills/tao-finetune-video-clip &&
rm -rf "$tmp"
```

For one project only, put the folder in `.claude/skills/tao-finetune-video-clip` inside it instead.

### Codex

Copy the skill folder to `~/.agents/skills/tao-finetune-video-clip` to use it in every repository:

```sh
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.agents/skills &&
rm -rf ~/.agents/skills/tao-finetune-video-clip &&
cp -R "$tmp/skills/tao-finetune-video-clip" ~/.agents/skills/tao-finetune-video-clip &&
rm -rf "$tmp"
```

For one repository only, put the folder in `.agents/skills/tao-finetune-video-clip` inside it instead.

### Cursor

Copy the skill folder to `~/.cursor/skills/tao-finetune-video-clip` to use it in every project:

```sh
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.cursor/skills &&
rm -rf ~/.cursor/skills/tao-finetune-video-clip &&
cp -R "$tmp/skills/tao-finetune-video-clip" ~/.cursor/skills/tao-finetune-video-clip &&
rm -rf "$tmp"
```

For one project only, put the folder in `.cursor/skills/tao-finetune-video-clip` inside it instead.

### Gemini CLI

Run in a terminal (add `--scope workspace` to install it for the current workspace only):

```sh
gemini skills install https://github.com/NVIDIA/skills.git --path skills/tao-finetune-video-clip
```

Or copy the folder to `~/.gemini/skills/tao-finetune-video-clip` yourself.

### VS Code

For GitHub Copilot in VS Code, copy the skill folder to `~/.copilot/skills/tao-finetune-video-clip`:

```sh
tmp=$(mktemp -d) &&
git clone --depth 1 https://github.com/NVIDIA/skills.git "$tmp" &&
mkdir -p ~/.copilot/skills &&
rm -rf ~/.copilot/skills/tao-finetune-video-clip &&
cp -R "$tmp/skills/tao-finetune-video-clip" ~/.copilot/skills/tao-finetune-video-clip &&
rm -rf "$tmp"
```

For one project only, put the folder in `.github/skills/tao-finetune-video-clip` inside it instead.

### Other agents

Get the folder:

```sh
git clone --depth 1 https://github.com/NVIDIA/skills.git
```

Then copy `skills/skills/tao-finetune-video-clip` into your agent's skills folder as `tao-finetune-video-clip`. Agents that support SKILL.md read one folder per skill.

## FAQ

### What does the environment need to run this skill?

Docker with nvidia-container-toolkit and the pinned TAO video_clip PyTorch and Deploy containers, or a local tao-pytorch checkout with a tao-cli virtualenv. MobileCLIP and InternVideo2 weights must be on disk for offline evaluation.

### What input format does it expect?

Metadata as a top-level JSON list of vadr1_chunks video records, each with an absolute video_path and nested caption chunks.

### Does it use AutoML?

No. AutoML is not packaged for this skill, so training always uses the direct video_clip train action.

### Which tools does the skill allow the agent to use?

It is declared to use Read and Bash.

### How do I install Tao finetune video clip in Claude Code?

Copy the skill folder from GitHub to ~/.claude/skills/tao-finetune-video-clip to use it in all your projects, or to .claude/skills/tao-finetune-video-clip in one project.

### Does Tao finetune video clip run scripts?

No. It is instructions only, with no scripts. This comes from a static check of the files on GitHub; nothing was run. Read the files before you install it.

### Which AI agents can use Tao finetune video clip?

Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Windsurf and Goose. Each reads skills in the SKILL.md format from its own folder; the install steps on this page list them.

### What are the best alternatives to Tao finetune video clip?

The closest alternatives to Tao finetune video clip are Skill creator, Rea tool design, Codebase design, Prototype and Brand guidelines.

## Where this comes from

- Description: from the skill's own SKILL.md.

## Alternatives

- [Skill creator](https://ai-powered.com/skills/skill-creator): Skill creator is a free agent skill for ai models and llm tooling.
- [Rea tool design](https://ai-powered.com/skills/rea-tool-design): Rea tool design is an open-source agent skill for design and media.
- [Codebase design](https://ai-powered.com/skills/codebase-design): Codebase design is an open-source agent skill for design and media.
- [Prototype](https://ai-powered.com/skills/prototype): Prototype is an open-source agent skill for developer tools.
- [Brand guidelines](https://ai-powered.com/skills/brand-guidelines): Brand guidelines is a free agent skill for design and media.
- [Theme factory](https://ai-powered.com/skills/theme-factory): Theme factory is a free agent skill for design and media.
- [PPTX](https://ai-powered.com/skills/pptx): PPTX is a free agent skill for files and documents.
- [Claude API](https://ai-powered.com/skills/claude-api): Claude API is a free agent skill for ai models and llm tooling.

---
Source: https://ai-powered.com/skills/tao-finetune-video-clip
