Appearance
The assistant
Describe the graphic you want, and the assistant builds it in your document: artwork, animation, data and 3D. You watch each step land on the stage, steer it in words, and undo it like any other change.
To drive the same work from your own Claude Code or Claude Desktop, see Use your own Claude with airZ Editor.
Ctrl+J, or the sparkle icon at the far right of the top bar. The icon turns blue while the window is open, and shows a dot while the assistant works.

Before the first use
The assistant needs two things, and says which is missing, with Check again and Settings…:
- The scene compiler, which comes with airZ Editor. If it is missing, the window says "The scene compiler is missing." — reinstall airZ Editor to get it back. See Settings › Scene compiler.
- A provider: your Claude subscription through Claude Code, or an API key. See Settings below.
Asking for something
Write what you want in the box — "a lower third for a news channel, name and role, with a blue accent bar that slides in" — and press Enter (Shift+Enter for a new line).
Pictures and fonts go with your next message: the paperclip, or drop them on the window, before your first message or later. PNG, JPEG, WebP, TTF and OTF. A picture is a reference to match, kept out of the graphic, unless you say otherwise: the line under the pictures reads "Pictures are references to match, kept out of the graphic. Switch one to In scene to place it." Each waiting picture has a label under its thumbnail; click it to switch:
| Label | What the assistant does with the picture |
|---|---|
| Reference ⇄ (a new picture) | Matches its layout, colours, type and mood, and compares its own preview against it. It never puts the picture in the graphic |
| In scene | Places it in the graphic: a logo, a photo. Click the label to switch before you send |

References stay with the session, so "closer to the reference" still works several messages later. A picture pasted rather than dropped is always a reference: to place it, drop the file instead.
While it works, each step is listed — "build the scene", "check the scene", "render a preview" — with a tick or a cross; click one to see what it did. Each build shows on the stage as it lands. Write while it works to steer it: Queue holds your message until the turn ends, Stop ends the turn.
| Switch | What it does |
|---|---|
| Live document | On, builds go into your open document. Off, they show on the stage as a preview, with Open in editor, Merge artboard and Back to document |
| Scripting | Lets it write scripts too. Off, it uses timelines, Animation States, data binding, constraints and listeners only |
Undoing it
There is no approval step: changes go straight into the document. At the end of a turn it says how many — "N changes to your document; one Ctrl+Z takes them all back." — and one Ctrl+Z does.
If you edit the document while it is building, it does not overwrite your work: "You edited the document since the last build, so this build was not put in it." Replace puts the build in anyway.
Sessions
Each conversation is a session, with the steps and files it made. The header's buttons: New session, Sessions… (open or delete an old one; double-click a title to rename it), Styles… (see below), Settings… and Close. Once a session has started, the window has two tabs: Chat, and Markup, which shows what the assistant wrote, with the lines it changed highlighted.
Styles
A look you want again — a channel's colours, its type, the mood of its pictures — can be kept as a style: reference pictures and a few words, under a name. Give the assistant the pictures, then ask, for example "save this style as Weather". Later, "use the Weather style" builds the next scene to match it, in any session. The assistant compares its preview with the style's pictures, and can take their colours as the document's named colours.

Styles… in the header lists them, each with its pictures, its words and "N pictures · saved … ago".
| To | Do |
|---|---|
| See a picture large | Click its thumbnail ("Show this picture large"); a click or Esc closes it |
| Rename a style | Double-click its name, type, Enter |
| Delete it | The delete button, then Delete at "Delete this style and its pictures?" (Keep leaves it) |
A style's pictures stay with the editor, never in a document: as the window says, "a style sets how a scene looks; its pictures are never placed in a document". With none saved, it says how to start: "No saved styles yet. Give the assistant reference pictures and ask it to save the style, for example "save this style as Weather"."
Limits
To keep a request from running away, the assistant pauses and asks before going on after many turns, after several failed builds in a row, or when a session has used its budget: "The assistant took N turns on this request; continue?". The limits are in Settings.
Settings
File › Settings… › Assistant, or Settings… in the window.
| Row | What it does |
|---|---|
| Provider | Claude Code (your Claude subscription), Anthropic API (a key with credits), or OpenAI-compatible (chat completions) |
| Model | Which model it uses. The default is right for most work |
| Claude Code | Whether Claude Code is installed and signed in, and Check again |
| API key | For the two key-based providers: paste it and Save, Test or Remove. It is kept in the Windows credential store, not in a file, and every request is billed to it |
| Scripting by default | Whether new sessions may write scripts |
| Turns per request, Failed builds before a pause, Tokens per session, Longest reply | When it pauses to ask |
The usage line at the foot of the window says what a session has used, and its cost where the provider says.
See also
- Settings · Keyboard shortcuts
- Versions: keep a version before a big request