The Graphics Generator

Q: My generated image isn't in the Assets Manager. Where is it?

It's held as a temporary generation (status deprecate) until you click Accept on it. Accepting promotes it to a permanent asset. Unaccepted generations are not kept.

How does the Graphics Generator work? Learn the Templates and AI Canvas tabs, the two-stage AI pipeline, and what Accept does to keep a generated image.

What you'll learn

What the Graphics Generator is and the two tabs it ships with.
How the AI Canvas two-stage pipeline turns a casual message into a brand-aware image.
What "Accept" does, and where the Graphics Generator is still rough.

Beta, up front

The Graphics Generator is in beta — it carries a BETA badge in the UI, and this article is honest about where it's rough. The pipeline works and produces images, but the flow has sharp edges (notably how a generated image is held as a temporary asset until you accept it, covered below). Treat it as a useful-but-young feature, not a finished one.

The concept: two ways to make a graphic

The Graphics Generator lives at /studio/graphics and has two tabs:

Templates — a layout-based generator. You pick a template (a spotlight card, a quote card, and so on), choose a format and theme, toggle whether to stamp the brand logo, and render a graphic from structured fields. No AI key needed. This is the predictable, repeatable path for things like social cards.
AI Canvas — a chat-style image generator. You describe an image in plain language and an AI pipeline produces it, using your active brand's visual identity as context. This is the open-ended path. It's BYOK-gated: the AI Canvas tab is disabled with a tooltip until you've set an AI key.

The rest of this article is about AI Canvas, since that's the part with a concept worth teaching. For the step-by-step, see Your First AI Canvas Session.

Why AI Canvas exists: brand context without the prompt-writing

The usual problem with a consumer image tool is brand consistency. Type "a business meeting, flat vector illustration" and you get arbitrary colors and a random style. To force your identity onto the output you'd have to hand-write a long prompt full of hex codes and stylistic constraints — every time. People skip that, and the brand's visuals drift into a mismatched pile.

AI Canvas addresses this by knowing which brand you're working in. Because the whole workspace is multi-brand, the generator already has access to the active brand's colors, logo, and the content, tags, or assets you've attached to the message. You write the casual version; the pipeline supplies the context. What exactly gets injected is covered in Referencing Your Brand in Prompts.

The two-stage pipeline

AI Canvas does not send your message straight to an image model. It runs two stages:

Stage 1 — the prompt engineer

Your short message (e.g. "draw a picture of a laptop") plus the active brand's visual context goes to a text model. That model's job is to rewrite your casual request into a detailed image prompt that honors the brand's colors and constraints. You can inspect the result: each generated message has a View prompt toggle that shows the engineered prompt that was actually used.

Stage 2 — the image model

The engineered prompt then goes to an image model, which renders the graphic and returns it to the chat. You choose which image model to use from the image-model selector in the AI Canvas toolbar; m18t doesn't bundle a model, so this runs on your provider key.

Both stages persist. Sessions are multi-turn — follow up with "make the laptop look more modern" and the pipeline keeps the prior context and produces a new iteration. The engineered prompt is stored alongside each generation.

Accept: from temporary to permanent

This is the rough edge worth understanding. A freshly generated image does not arrive as a finished asset. It lands with the status deprecate — an intermediate, "not yet kept" state — and shows an Accept button. Most generations are throwaway iterations, so nothing becomes permanent until you decide it should.

Click Accept on a generation you want to keep. That promotes the image to a permanent asset in your Assets Manager, where it's available to attach to content and elsewhere. After accepting, the button is replaced by an Accepted badge. If you don't accept a generation, it stays in the temporary state — don't rely on it as a permanent file.

The BYOK economics

Image generation costs money to run, and m18t doesn't mark it up. As covered in Why m18t uses BYOK, you supply your own AI provider key via the secret vault and pay the provider directly. No bundled tokens, no platform quota on generations — and no AI key means the AI Canvas tab stays disabled.

What's next

FAQ

Why is the AI Canvas tab greyed out? No AI key is set. AI Canvas is BYOK-gated and disabled with a tooltip until you add an OpenAI, Gemini, or Anthropic key to the vault. The Templates tab works without a key.

My generated image isn't in the Assets Manager. Where is it? It's held as a temporary generation (status deprecate) until you click Accept on it. Accepting promotes it to a permanent asset. Unaccepted generations are not kept.

Which image model does AI Canvas use? Whichever you pick from the image-model selector in the toolbar. m18t doesn't bundle a model — you bring your own provider key.

Do I need a key for the Templates tab? No. Templates is a layout-based renderer and needs no AI key. Only AI Canvas requires one.

Is the Graphics Generator finished? No — it's in beta. The pipeline produces images, but the accept-to-keep flow and light safety filtering are known rough edges. Use it accordingly.