Use Video with your AI

Video has no built-in AI. Your own AI, Claude, ChatGPT or Grok, works with Video through Geenius OS: it reads a template's schema, drafts a video, edits scene by scene, checks the quality bar and lays sound on the timeline. A person renders, in a browser.

Connect in three steps#

In Geenius OS, an owner or admin adds Video to the workspace (Apps, Add to this workspace). Each workspace keeps its own apps.
In your AI, add the Geenius OS connector and sign in with your Geenius OS account. In Claude: Settings, Connectors, Add custom connector. In ChatGPT or Grok: add a remote MCP server. Geenius OS shows you the address.
Ask your AI to use Video. It now has a tool called geenius_video.

How a call looks#

The tool takes a tool (the area), an action (the verb) and args.

List the templates

{ "tool": "template", "action": "list", "args": {} }

Every call acts in the Video workspace of your Geenius OS workspace, after the plan check.

Your first ask#

A first ask

Use Geenius OS and my Video account. Make release notes for v2.4 from this changelog: <paste>. Use the release-notes template at 1080p, check it with lint, and give me the link so I can render it.

Your AI reads the template's schema and a valid sample, creates the draft, fixes what lint reports and gives you the link. You open it, watch the preview and press Render.

The tools#

Tool	What it is for	Actions
`video`	Videos and scenes	`list` `read` `review` `create` `update` `delete` `duplicate` `scene_list` `scene_read` `scene_add` `scene_update` `scene_remove` `scene_move` `scene_duplicate` `scene_copy_from` `chapters` `split` `component_list` `component_read` `ui_list` `ui_read` `lint` `audio_library` `audio_track_add` `audio_track_update` `audio_track_remove`
`template`	Templates	`list` `read` `sizes` `themes` `features` `films` `pack_list` `pack_read` `create` `from_video` `update` `param_add` `param_set` `param_remove` `scene_add` `scene_update` `scene_remove` `scene_move` `sample_set` `duplicate` `version` `versions` `diff` `upgrade` `share` `archive` `restore` `delete` `usage` `preview` `validate` `export` `import`
`collection`	Collections	`list` `read` `create` `update` `delete` `add` `remove` `reorder` `duplicate` `retheme` `resize`
`asset`	Pictures	`list` `upload` `import_url` `update` `library` `import`
`snippet`	Snippets	`list` `read` `create` `update` `insert` `delete`

Ask for the actions of an area with action: "help", and for one action's arguments with action: "help" and of.

Habits that work#

Read the schema first. template read gives a valid sample. Start from it and change values.
Edit scene by scene. scene_add, scene_update and the rest are one atomic edit each, and every write validates the whole spec and names the scene and field when something is wrong.
Lint before you hand over. lint lists the quality warnings in plain words.
Think in chapters. One render pass is up to 4 minutes. For a longer film, make chapters and split them.
Pictures by reference. Upload with asset and use storage:<id>; do not paste large data URLs into a spec.

What your AI cannot do#

It cannot render, and it cannot change the file of an imported film. It cannot make footage: Video draws from data, it does not generate pictures or clips.