Your First AI Canvas Session
What you'll learn
- How to start an AI Canvas session and generate your first image.
- How to attach brand context and iterate on a result.
- How to keep an image you like — and what happens to the ones you don't.
Before you start: you need an AI key
AI Canvas is BYOK. If you haven't added an AI provider key, the AI Canvas tab is disabled with a tooltip and you can't generate. Add an OpenAI, Gemini, or Anthropic key to the secret vault first. (The Templates tab works without a key — but this guide is about AI Canvas.)
For the why behind the two-stage pipeline, read The Graphics Generator first. This is the hands-on version.
Generate your first image
- Go to Graphics Generator at
/studio/graphicsand open the AI Canvas tab. - Click New Session to start a chat session.
- In the input box (placeholder: "Describe your image..."), type a plain-language description — e.g. "a server rack standing alone in an empty desert at dusk."
- Press Enter or click the send button.
m18t rewrites your message into a detailed, brand-aware prompt (stage one), then sends that to your selected image model (stage two). The generated image appears in the chat. You can expand View prompt under a result to see the engineered prompt that was actually used.
Make sure the right brand is selected in the workspace before you generate — the active brand's colors and logo are part of the context. See Referencing Your Brand in Prompts.
Add context for better results
Below the chat is a context bar. Before sending a message you can:
- Attach an Asset, Content, or Tags with the buttons in the context bar, so the generation is informed by an existing image, a content piece, or your tags.
- Toggle brand Colors and Logo so the pipeline pulls the active brand's palette and logo into the prompt.
The empty session even reminds you: "Attach assets, content, or brand context for better results."
Iterate
Sessions are multi-turn. If a result is close but not right, send a follow-up in the same session — "make it warmer," "add a sunrise," "remove the text." The pipeline keeps the prior context and produces a new generation, so you refine rather than restart.
You can also pick a different image model from the model selector in the toolbar between generations.
Keep the ones you want
A generated image is not saved as a permanent asset automatically. It lands in a temporary state and shows an Accept button.
- Click Accept on a generation you want to keep. It's promoted to a permanent asset in your Assets Manager, and the button becomes an Accepted badge.
- Generations you don't accept stay temporary — don't treat an un-accepted image as a saved file.
Once accepted, the image is available to attach to content and use anywhere assets are used.
Common variations
Switching brands mid-work. The session belongs to the brand you generated under. Switch brands and you're in that brand's Graphics Generator with its own sessions and visual identity.
Templates instead of AI. If you want a predictable, repeatable layout (a social card, a quote card) rather than open-ended generation, use the Templates tab — no AI key required.
What's next
FAQ
Why is the AI Canvas tab greyed out? No AI key is configured. Add one to the vault and the tab enables. The Templates tab works without a key.
Where did my generated image go? It's a temporary generation until you click Accept on it. Accepting promotes it to a permanent asset in the Assets Manager. Un-accepted generations are not kept.
How do I change a generated image? Send a follow-up message in the same session describing the change. The pipeline keeps the previous context and produces a new version.
Can I choose which image model to use? Yes — pick one from the model selector in the AI Canvas toolbar. m18t doesn't bundle a model; it uses your provider key.
Is AI Canvas reliable? It's in beta. It generates images, but expect rough edges — especially the accept-to-keep step. See The Graphics Generator.