How do I turn a beat into an illustration?

[MarcoDroll]

A walkthrough of the Illustrate this moment feature — what it needs, how the prompt gets written for you, and how to make a character look the same across several pictures.

[MarcoDroll]

Illustrate this moment takes a beat you've already generated and turns it into a picture placed directly in your prose. You don't write an image prompt — it reads the beat and writes one for you, which you then edit.

Before you start: you need an image backend

• Hosted, on a paid plan or the 7-day trial — nothing to set up. An included image model is ready to go, and renders draw on your monthly Included-AI budget. (The trial runs at the Premium tier.)
• Your own key — go to Settings → AI Models, enable fal.ai, OpenRouter or Replicate, and paste your key. Works on the hosted app and self-hosted.
• Self-hosting — Included AI is hosted-only, so your own key is the only route here.

If none of those is in place, the Illustrate button is visible but disabled, and hovering it tells you why.

Step 1 — Generate the beat first

Illustrate works from what the beat actually says, so generate the prose before illustrating it. The button stays disabled while a beat is still streaming, and while it has neither prose nor a prompt to read.

Step 2 — Press Illustrate this moment

It's the tinted button leading the beat's header row.

Step 3 — Check the prompt — it's already written

The app reads the beat's prose, your staging notes, and the codex entries relevant to that beat, then writes the image prompt. It comes back split into separate fields rather than one long line:

• one field per character, each labelled with that character's codex name;
• Anyone else — people in the shot with no codex entry (a crowd, a guard, a passer-by);
• The shot — place, hour, weather, light, mood, framing.

Every field is editable — type over anything you don't like. Splitting it this way means you can fix how one person looks without rewriting the whole description. Below the fields you'll see the Style and Genre lines that get applied at render time; those aren't editable text, they're settings (see step 5).

If the app can't write a prompt, it says so and leaves the field empty rather than filling it with something misleading — press Try again, or just describe the moment yourself and carry on.

Step 4 — Who's in this picture

Under the prompt is a switch per character. Turning someone on or off rewrites the prompt for the new cast, keeping any wording you typed by hand. Turning a character off only excludes them — switch them back on and their description returns exactly as it was, your edits included.

Step 5 — Render settings

• Image source — Included or Your key, if you have both available.
• Image model — on the own-key path you pick the model; the list shows what your configured providers offer.
• Shape — 16:9, 1:1 or 3:4. Some models render at a fixed shape and ignore this; when that happens the setting greys out and names the model.
• Style — Cinematic, Photorealistic, Illustrated or Minimalist. It defaults to your story's cover style, so a story's pictures match its cover with no setup, and changing it here sticks for that story.

Step 6 — Generate, then choose

Each render appears as a preview. Up to four are kept side by side (the oldest drops off), so you can compare and pick. Each finished preview tells you whether a character portrait guided it.

⚠️ Every render counts against your budget whether or not you insert it — that's true of the included path; on your own key, it's your provider you're paying.

Nothing is saved anywhere until you press Insert. Cancelling discards every preview.

Step 7 — Insert

The picture lands in your prose directly after the beat, with the cursor placed underneath it so you can keep writing. The prompt is stored as the image's alt text. Later pictures for the same beat are appended after the earlier ones, so nothing you've written gets pushed around.

Removing the picture from your prose also cleans up the stored file, so deleted images don't linger in your story's media.

Making a character look like themselves

By default everyone in the picture is described in words, which means the same character can come out looking different from one render to the next. To pin their appearance, a codex portrait can be sent to the image model as a reference:

Three things have to line up:

1. the character has a portrait on their codex entry;
2. you're on Your key with a fal.ai key configured;
3. you've picked one of the character-reference models (they're grouped at the top of the model list).

The Included image model can't accept portrait photos — that's the main reason to bring your own fal.ai key for this feature. Each reference model also takes a limited number of photos at once; anyone beyond that limit is described in words instead. Whenever a reference can't run, the dialog says which of the three reasons applies, and every finished picture records whether it used one.

Where the related settings live

• Settings → AI Prompts → Illustration Prompts — which model writes the prompt (not which renders the image), plus a custom instruction if you want to steer every prompt.
• Genre is set when you create a story, and it shapes the illustrations generated from its beats.

If something goes wrong

• "The image prompt was rejected" — the included path runs a content check; adjust the description. A refusal costs you no budget. (The own-key path has no such check — that's your provider's call.)
• "Your monthly AI budget is used up" — it resets at the start of your next billing cycle.
• "Too many image requests just now" — your prompt is fine, wait a moment and retry.