video-understanding
Guide video upload, captioning, action recognition, dense captioning, and TikTok-style clip generation.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- video-understanding
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guide video upload, captioning, action recognition, dense captioning, and TikTok-style clip generation.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/troubleshooting.md3.0 KB
- 参考文档references/video-tool-reference.md4.3 KB
- 参考文档references/video-workflows.md3.8 KB
03 SKILL.md 正文
video-understanding
Use this sub-skill for local video requests such as caption this video, action recognition, dense caption, GenerateTikTokVideo, cut video to TikTok, or uploading a clip for later discussion.
Read references/video-workflows.md when choosing between upload, caption, action, dense, or TikTok flows. Read references/video-tool-reference.md when you need the exact input/output contract, timeline format, or checkpoint map for a specific video tool. Read references/troubleshooting.md when a checkpoint, CUDA/VRAM limit, ffmpeg, Bark, or OpenAI prerequisite is missing. Run scripts/validate_video_plan.py to statically check a video path, requested tool names, and TikTok prerequisites before any runtime plan.
Scope
- In scope: video upload state, optional pre-captioning, Tag2Text captions/tags, InternVideo action recognition, GRiT dense captions, TikTok-style clip generation, and artifact naming.
- Out of scope: app launch/load flags, HTTPS/service deployment, and non-video modalities.
Routing hints
- If the request is about app launch, load strings, tabs, or other service flags, hand it to
../app-deployment/SKILL.md. - If the request is image, audio, or DragGAN work, use the sibling modality skill instead of this one.
Expected outputs
- Uploaded videos become renamed local copies in the app's working media area.
- Captioning returns caption text plus tag lists and framewise detail strings.
- Action recognition returns a Kinetics class label.
- Dense captioning returns timestamped object-name descriptions.
- TikTok generation returns a new mp4 clip plus temporary clip and audio artifacts.
Guards
- Do not claim a video result without its required checkpoint, codec, or external tool.
- Do not promise TikTok generation when OpenAI, ffmpeg, or Bark is unavailable.
- Do not import or invoke heavy models from this skill; use the runtime tool plan only.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.