visual-notes — Claude Code skill for portrait video editing
This Claude Code skill shifts the image above an isolated speaker from a full view to a close-up detail. Run /visual-notes to apply the treatment to a portrait clip.
MIT ★ 94 updated 2026-10-30
/plugin marketplace add artemnovitckii/claude-editing-skills
/plugin install editing-skills@claude-editing-skillsWhat visual-notes does
The skill builds a portrait video where the image above an isolated speaker starts as a whole photograph, then moves in on a chosen detail. It includes a 4.2-second starter at 1080x1920 and 30fps. The user supplies the topic; the sample speaker’s words do not determine it.
Run build.py with a path to a spec file. Without one, the script recreates a silent sample in .work/; --prepare-only creates an editable composition without rendering. Requirements are Python 3.9+, ffmpeg, ffprobe, Node.js 22+, and npm. The first build fetches GSAP 3.15.0, and HyperFrames 0.8.214 renders the video. No API key or transcript is needed.
Replacing the presenter requires a frame-matched alpha cutout at the same dimensions; kit/cutout.py can produce one. The build helper does not segment footage automatically. Longer edits need an extended HTML timeline and source windows, since changing the duration alone will not extend the edit.
What it helps with
- create a portrait clip that moves from a full image to a close-up
- change the photograph and adjust its crop anchor
- prepare an editable composition without rendering
- validate and decode the rendered MP4
Example prompts for the agent
- Apply
/visual-notesto a portrait clip about the new product. - Build a portrait edit from
spec.json, moving from the full photograph to one detail. - Prepare an editable composition with
--prepare-onlyand do not render it.
How to install visual-notes
1Plugin marketplace
/plugin marketplace add artemnovitckii/claude-editing-skills/plugin install editing-skills@claude-editing-skillsThis installs the whole editing-skills plugin with all its skills
2Skills CLI (Vercel)
npx skills add artemnovitckii/claude-editing-skills --skill visual-notes -a claude-code3Manually: put the files in place
~/.claude/skills/visual-notes/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/visual-notes/
1Skills CLI (Vercel)
npx skills add artemnovitckii/claude-editing-skills --skill visual-notes -a codex2Manually: put the files in place
.agents/skills/visual-notes/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add artemnovitckii/claude-editing-skills --skill visual-notes -a cursor2Manually: put the files in place
.cursor/skills/visual-notes/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install artemnovitckii/claude-editing-skills skills/visual-notes2Manually: put the files in place
.github/skills/visual-notes/SKILL.md1Skills CLI (Vercel)
npx skills add artemnovitckii/claude-editing-skills --skill visual-notes -a gemini-cli2Platform CLI
gemini skills install https://github.com/artemnovitckii/claude-editing-skills --path skills/visual-notesManually: put the files in place
.devin/skills/visual-notes/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about visual-notes
How do I install visual-notes in Claude Code?
Add the marketplace with /plugin marketplace add artemnovitckii/claude-editing-skills, then install the plugin with /plugin install editing-skills@claude-editing-skills.
What does visual-notes do?
It builds a portrait video that changes the image above an isolated speaker from a full view to a close-up of a selected detail.
What does visual-notes need to run?
It requires Python 3.9+, ffmpeg, ffprobe, Node.js 22+, and npm. The first build downloads GSAP 3.15.0, and rendering uses HyperFrames 0.8.214; no API key or transcript is required.
Source text
SKILL.md · MIT
--- name: visual-notes description: An image above the isolated speaker changes from a complete view to a deliberate close-up detail. Use for /visual-notes or to apply this specific portrait-video treatment. --- # /visual-notes Apply this treatment to a portrait clip using the included 4.2-second, 1080x1920, 30fps starter. The style and images are independent of the sample speaker’s words; follow the user’s chosen topic. [See the reviewed example](../../docs/style-showcase/visual-notes.mp4). ## Use Resolve this skill folder from the base directory shown when the skill loads. Its `build.py` resolves the shared kit through that real path, including when installed as a symlink. Run: ```bash python3 /path/to/skills/visual-notes/build.py /path/to/spec.json ``` With no spec, the command reproduces the included silent sample into the repository’s `.work/` directory. `--prepare-only` builds an editable composition without rendering. Python 3.9+, ffmpeg/ffprobe and Node.js 22+/npm are required. The first build downloads pinned GSAP 3.15.0 into the ignored vendor cache; rendering uses HyperFrames 0.8.214. No transcript or API key is needed for these short templates. [Spec and custom-footage instructions](../../docs/style-showcase/README.md). ## Adapt the treatment Show the whole photograph first, then expose a specific detail at 1.5s. Re-anchor the crop inside `#iris` when changing the photograph; the example focuses on the pixelated burger detail. Update `#tag`, `#callout` and both initial and GSAP-updated text. The magnification should reveal something real in the image. Asset keys: `explainer_cut.webm`, `pixel-burger.jpg`. Full-resolution example images and source credits are in [assets/style-showcase](../../assets/style-showcase/README.md). Supply a same-size, frame-matched alpha cutout when changing the presenter. The existing `kit/cutout.py` can create it; see the shared guide. Inspect hair and hands over both bright and dark imagery. The build helper does not perform segmentation automatically. Keep the presenter visible from the first result frame. Use sharp photographs and clean type. Essential text and the face should clear platform overlays; let decorative imagery bleed rather than shrinking the entire composition. The sample’s captions and image anchors are starting points, not automatic tracking. ## Review and deliver The helper validates media dimensions/duration, prepares the composition, runs the renderer check, renders and fully decodes the MP4. Inspect the first frame, the main change and the ending. Check crop anchors, text readability and matching source/cutout timing. The default is silent; preserve or add speech, SFX or music only when the user requests them. Longer edits require extending the HTML timeline and source windows; changing only the duration is insufficient. These are local Claude Code/Codex editing workflows. The ChatGPT conversation in the showcase is an authored visual, not a built-in ChatGPT slash-command integration. Third-party image rights are documented separately from the code license in the asset directory.
© Artem Novitckii · MIT · artemnovitckii/claude-editing-skills