AI Studio overview

AI Studio gives a prompt, file, URL, or group of canvas objects a useful next form. That can mean generating an artifact, saving source material, or opening a focused question flow.

The useful first question is not “What can AI make?” It is “What should the next person be able to open, inspect, or change?” This overview maps that decision to the right guide.

Start with the next action

Choose Auto when you can describe the outcome but do not want to decide the output or action. Choose a named type when the deliverable is clear. Create content with AI Studio covers prompts and context. Use AI Studio in a canvas covers panel location, access, placeholders, and cancellation.

AI Studio can use prompt text, selected canvas objects, completed uploads, safe HTTPS URLs, and voice input. Keep the context focused. Uploads still processing are not included, and failed uploads need to be removed or retried.

Choose the result people need next

The subject does not decide the type. A written brief about a landing page is a Doc. A page people should open and review is a Web page.

See three generation flows

Image generation starts with the references and instructions already on the canvas, then places the finished visual beside that context for review.

Table generation turns a request into an editable grid, so the structure can be checked and corrected without leaving the canvas.

Web page generation carries the same source context into a live preview that can be opened, tested, and shared.

Know what Auto decides

Auto interprets both the request and the source context. With a prompt alone, it resolves the request to a concrete generation type. With completed attachments or prompt URLs, it first decides what you are trying to do:

• A request to keep the source creates source objects on the canvas.
• A question about the source saves it and opens Object Chat for a focused conversation.
• A request to transform the source starts generation and creates a new result.

This prevents “What does this say?” from producing an unnecessary artifact. Auto is an intent choice, not just another output type. Choose directly when the action or deliverable is already clear.

Ask about an item instead of creating one

When the job begins and ends with one existing item, use Object Chat or Ask ALLO. It is built for understanding, summarizing, and critiquing that item, and it can edit supported items. It gives “this” an explicit meaning and keeps the conversation attached to its source.

Use AI Studio for a new canvas result or context from several objects. Use Object Chat when one item is already the center of the work. Human feedback still belongs in comments.

Review the draft before it becomes shared work

Generated output appears on the canvas near the work, but proximity is not approval. Check facts, dates, numbers, names, source completeness, permissions, and sensitive content before anyone relies on it. A polished sentence can still be wrong, and a clean table can still contain an invented cell.

Review AI output and share web pages explains placeholders, previews, AI badges, output-specific checks, and webpage sharing. Use it before a client handoff or decision record. If a result is close, edit it instead of asking AI to guess again.

Plan for credits and recovery

AI generation, Object Chat, AI edits, and voice transcription use workspace AI credits. Understand AI credits explains what spends credits, where admins see the balance, and what happens when credits run out. Purchases and Auto-reload are admin responsibilities.

When a result is wrong, identify whether the cause was the source, prompt, output choice, or an incomplete attachment. Add the missing audience, format, constraint, or source before regenerating. Do not submit duplicates while a placeholder is still processing because longer image, audio, file-heavy, and webpage jobs continue in the background.

When an AI result is not right covers wrong, stuck, incomplete, and credit-blocked results. It separates prompt problems from attachment, permission, processing, and billing issues so the next action fixes the cause instead of repeating the spend.