Explainer editor help

Your first video in 3 steps

1. Open the editor and write your topic in one sentence, or click “Try with an example” (no AI budget spent).
2. Review the scenes: click an element to move it, change a word of the narration, or ask the Director for a tweak.
3. Start the render (⌘↵): the MP4 shows up in your list, ready to download.

Start the guided tour 5 tips inside the editor, skippable at any time.

Keyboard shortcuts

In the editor, “?” opens this list. Single-key shortcuts don't fire while typing and can be turned off in that same window.

Concepts

Scene

A scene is one narration sentence and the elements that illustrate it. The video is the sequence of your scenes in the timeline.

What the AI does: it splits your topic into scenes and estimates their length.

Undo: ⌘Z, or reorder and delete scenes in the timeline.

Element and library

An element is an animated drawing (2D pixel art or 3D object) placed in a scene. The library holds about 11,000 of them, filed by theme and searchable by meaning (“/”).

What the AI does: it suggests elements matching the meaning of each sentence and times them to the words.

Undo: ⌘Z, or delete the element (⌫).

Director

The Director is a copilot you talk to (“make scene 3 more dramatic”). It proposes script changes, shown as a diff before they apply.

What the AI does: each proposal is summed up in one sentence; you accept or decline.

Undo: “Decline”, or ⌘Z after accepting; “Stop” cuts a running request.

Polish (critique)

Polish makes the AI review its own frames: overlaps, empty areas, off-topic elements. Each scene gets a score and suggested fixes.

What the AI does: fixes arrive as a Director proposal.

Undo: decline the proposal, or ⌘Z; a version is kept in the history.

Forged elements

When an element is missing, Claude draws it from your description, renders it and reviews it. It goes to “My elements”, private, and you can propose it for publication.

What the AI does: drawing, render, vision review (2 fixes at most), cost shown.

Undo: “Discard” moves it to the trash for 30 days, “Undo” restores it.

Narrate and imported audio

Narrate into the mic (one or more takes) or import an audio file: the narration is transcribed word by word and the video builds on your voice. You can re-record a single sentence.

What the AI does: transcription, scene split and illustration timed to your words; optional audio cleanup.

Undo: the original audio is kept: cleanup and re-recording can be undone (⌘Z).

Variants

“3 variants” proposes three takes on a scene, side by side, with their cost shown before you start. You choose instead of building.

What the AI does: three different proposals for the same sentence.

Undo: picking a variant can be undone (⌘Z); the previous scene stays in the history.

Versions (history)

Every generation, accepted proposal, Polish and import adds a version, with a thumbnail and a readable diff. You can compare two versions and go back to one.

What the AI does: comments left on the timeline (C key) can be resolved by the Director.

Undo: going back keeps the current version; “Undo” reverts the restore.

Taste

The AI remembers your choices (accepted or declined proposals, elements kept or discarded, pacing) and applies them to the next videos. The “What the AI knows about my taste” page shows everything it keeps.

What the AI does: the Director says which preference it applied.

Undo: delete a preference, turn off “Learn from my choices”, or “Forget everything”.

Brand

The brand kit (logo, colors, font, tone, intro and outro, watermark) is applied by the engine to every render. Set it in Settings > My brand.

What the AI does: tone and colors guide the composition.

Undo: turn the brand off for one video, or delete the kit.

Templates

A template is a reusable script with parameters such as {{sujet}}. “Fixed” fills the parameters without AI; “Recompose” lets the AI rewrite from the template.

What the AI does: in Recompose mode only, with the estimated cost first.

Undo: the produced script edits and undoes like any other; delete a template in “My templates”.

Agents

An external agent (Claude Code, Claude Desktop, a script) can produce a video end to end through MCP or the API, with a key scoped “read” or “production”. Produced videos carry the key's name in your list.

What the AI does: composition, check and render with no human, live progress.

Undo: cancel a running production; revoke the key at any time.

Connect your agent

1. Create a key in Settings > Integrations: “read” to look, “production” to produce. The secret is shown only once.
2. Plug in the MCP server (claude mcp add … /mcp with the Authorization header) or call POST /api/agent/produce.
3. Always start with dry_run: estimated cost, nothing billed. Follow the production (SSE or signed webhook) and cancel it if needed.

Create an API key Interactive API reference · llms-api.txt

Step-by-step guide with real outputs (MCP, dry_run, signed webhook): docs/agents.md.