Referencing Your Brand in Prompts

What brand context can AI Canvas inject into an image prompt? Learn which context is opt-in via toggles and attachments, and how session history carries it.

What you'll learn

  • Exactly what brand context AI Canvas can inject into an image prompt.
  • Which context is opt-in via toggles and attachments, and which comes for free.
  • How conversation history carries context across a multi-turn session.

The concept: context the model can use

When you generate an image in AI Canvas, the first stage of the pipeline is a text model that rewrites your casual message into a detailed image prompt. The quality of that rewrite depends on what context it has. m18t can hand the model several pieces of context drawn from your active brand and from the things you attach — so you describe the image loosely and the pipeline fills in the brand-specific detail.

This article spells out precisely what gets injected, so you know what to set up and what to expect. It is honest about the current behavior: most of this context is opt-in per message, not bundled into every generation automatically.

What can be injected

When you generate, AI Canvas assembles a brand-context block from these sources:

  1. Brand name and primary color — included when you turn on the Colors toggle in the context bar. The block names the active brand and its primary brand color.
  2. Brand logo — included when you turn on the Logo toggle, and the active brand actually has a logo set. The logo URL is added to the context.
  3. Referenced content — if you attach a content piece, its title (and excerpt, if it has one) is added, so the image relates to a specific post.
  4. Tags — if you attach tags, their names are added as themes for the image.
  5. Referenced assets — if you attach existing assets, their alt text (or filename) is added as reference descriptions.
  6. Prior conversation — the messages already in the session are passed along as history, so a follow-up like "make it warmer" is understood in context.

If you attach nothing and toggle nothing, the context block is essentially empty and the model works from your message alone. That's the key honest point: in AI Canvas today, brand colors and logo are toggles you switch on per message — they aren't forced into every generation automatically. Turn them on when you want on-brand output.

How to set it up

  1. Make sure the brand you want is selected in the workspace. Its name, color, and logo come from the brand's settings — fill those in under your brand's configuration so there's something to inject.
  2. In the AI Canvas context bar, turn on Colors and Logo to pull the active brand's palette and logo into the prompt.
  3. Use the Asset, Content, and Tags buttons in the context bar to attach anything you want the image to reference.
  4. Send your message. Expand View prompt on the result to see how the context shaped the engineered prompt.

Common variations

No logo injected even with the toggle on. The logo is only added if the active brand has a logo URL set. If your brand has no logo, set one in the brand's configuration first.

A plain, off-brand result. Likely nothing was attached and the Colors/Logo toggles were off, so the model had no brand context. Turn the toggles on and attach relevant content or tags, then regenerate.

Refining within a session. Because prior messages travel as history, you don't re-state context on every turn. Keep iterating in the same session and the pipeline remembers what came before.

What's next

FAQ

Does AI Canvas automatically use my brand colors? Not automatically — you turn on the Colors toggle in the context bar to inject the brand name and primary color into the prompt. Same for the Logo toggle.

Why didn't my logo get used even with the Logo toggle on? The logo is only injected if the active brand has a logo set. Add one in the brand's configuration, then generate again.

What exactly does the model see about my brand? Depending on your toggles and attachments: the brand name and primary color, the logo URL, an attached content piece's title and excerpt, attached tag names, attached assets' alt text or filenames, and the prior conversation in the session.

Do I have to re-attach context for every follow-up? No. The session's prior messages are passed as history, so follow-ups are understood in context. Re-attach only when you want to add something new.

Can I see the prompt the model actually used? Yes — expand View prompt under a generated result to see the engineered prompt.

Was this article helpful?