Use AI Studio

AI Studio turns prompts, selected canvas objects, uploaded files, pasted URLs, and voice input into canvas objects you can move, edit, comment on, ask about, and share. It makes the output next to the work it came from and the people who will review it, instead of in a separate AI tab.

Treat every result as a draft. AI Studio is good at getting the first shape onto the canvas; you still own the facts, tone, dates, source permissions, and final handoff.

This is the reference for AI Studio. For the step-by-step canvas-entry flow, see Use AI Studio in a canvas.

Who can use AI Studio

AI Studio creates canvas content, so it is available to workspace members who can edit the canvas. On desktop it appears as the canvas AI panel; on supported mobile or narrow screens it uses the canvas toolbar panel.

Credits are checked at submit time: AI work starts only after the workspace AI credit gate passes. Workspace admins buy credits in Billing; other members need an admin to add them.

Choose the output type

Auto is the safest starting point — it resolves the output type from the prompt and context, and with completed files or URL tiles it also decides whether to save source material, open Object Chat, or generate.

For images, Count goes up to 10. Leave Model, Style, and Ratio on Auto unless the result must match a specific production format (a vertical social image, a square concept tile, a wider presentation visual). Auto is the planned path that lets the backend pick from the prompt and references; change the controls only when format actually matters.

Give AI Studio the right context

The best prompt gives AI Studio the material it needs, not the most words.

New placeholders anchor near the visible center of your viewport, not on the selection, so generated work appears where you can see, move, or cancel it. On a freeform whiteboard, smart arrows from a source object to a result let AI Studio and Object Chat treat the source as prior context.

Attachments

AI Studio accepts up to 10 attachments per draft (images, PDF, Office and HWP, Markdown, text and code, CSV, JSON, SRT, and spreadsheets), reading each for context up to a 20 MB per-file cap. Split oversized source material first.

Completed uploads are included in the request; in-progress uploads are not; failed uploads stay visible with an error so you can remove and retry. Draft recovery stores completed upload metadata, prompt text, selected context, and options — but not raw in-progress file bytes, so reattach any file that was still uploading when you refreshed.

When Auto sees completed attachments or prompt URLs, it runs an intent check before generating:

So "what does this say?" on a pasted URL gives you a source object plus Object Chat, while "turn this into a client-ready launch brief" gives you a generated Doc.

Write prompts that leave less cleanup

Give the model the decision shape you want, not just the topic — audience, format, source objects, and constraints. For example: Turn the selected notes into a one-page project brief for a product lead, with headings, risks, decisions, and next steps.

If a result is bland, do not keep regenerating the same vague request. Add the missing audience, format, source object, or constraint — manual editing is often cheaper than asking AI to guess again.

Voice input is prompt writing, not generation: it checks credits before the browser microphone starts, then records, transcribes, and appends the transcript. Empty or no-speech recordings are treated as empty input; microphone denial, missing devices, and transcription failures surface as prompt-composer errors. Skim the transcript before submitting, because a small transcription mistake becomes a large mistake in the object.

Watch the placeholder, then review the object

After AI Studio accepts a request, work continues in the background and the placeholder is the live sign of progress. While it processes, output actions — edit, preview, download, duplicate, copy, lock, thumbnail — stay blocked; selection and deletion are the safe controls.

Deleting the processing placeholder is the cancel gesture, and deleting the canvas cancels matching running tasks. If generation already finished or credits were already consumed, deleting the final object does not undo the spend. Longer file-heavy, audio, image, and web page jobs continue server-side after you leave, so return to the canvas to review rather than resubmitting the same expensive prompt.

AI Studio vs Object Chat

Use AI Studio when the next step is to create or save something on the canvas. Use Object Chat when the next step is to ask about one existing object or keep working on one result — it explains, rewrites, critiques, or edits an artifact after it exists. Generated and AI-edited objects can show an AI badge as their object-level AI entry point, which is separate from human comments. Before using AI output in client-facing work, read Review AI output and share web pages.

Credits are workspace spend

AI credits belong to the workspace. The actor is recorded for audit, but the balance is checked at the workspace level before provider calls. AI Studio generation, Object Chat messages, recommendation buttons, and voice transcription all pass through that gate.

Admins recover by buying credits in Billing; non-admin members, external collaborators, and guests cannot buy credits from the blocked flow and need a workspace admin. Successful AI work can spend credits even if the output is later wrong, deleted, or needs cleanup, so check that the prompt and source context are worth the spend before a large image, audio, or web page request. For details, read Understand AI credits and AI credit billing.

Troubleshooting

When you contact support, send the workspace name, canvas link, output type, prompt summary, attachment types, approximate time, and whether credits were blocked or spent. Do not send passwords, API keys, private customer data, or unnecessary source files.