Images — generate and refine

Ask for a picture in plain language and Vancetope generates one, right in the chat. No separate tool, no prompt syntax to learn — describe what you want, and refine it by asking for changes. Every image is saved as a document in the project, so you can open, download or reuse it later.

Behind the scenes this is Fenchurch, Vancetope’s image service. You never pick a model or endpoint — the agent calls it for you and streams progress while it works.

Table of contents

1. Ask for an image
2. What you get back
3. Where to go next

---

Ask for an image

Just say what you want. The agent recognises the intent, calls the image tool, and the finished picture lands in the conversation with a title — here, a painterly mountain lake at sunrise.

Generate & refine — 2 steps

Step 1Describe the picture. The image renders inline with a title; the live-progress panel shows the image_generate call — "Generated in about 8 seconds with the fast model."

Step 2Ask for changes in the same breath — "more vibrant, add a small rowboat" — and it generates an updated version. Iterate until it's right.

What you get back

Each image is a real document in the project, not a throwaway. The card in the chat carries an open (↗) and a download (↓) affordance, and the file lives under the project’s images/ folder — ready to drop into a workpage, a canvas, or an export.

You don’t choose a model. The operator wires image generation once (a provider key plus the default:image / default:image-high aliases); from then on everyone just asks. Two quality tiers exist — a fast, cheap default and a higher-quality one for final assets — and the agent picks based on what you ask for.

Where to go next

• The service behind it: Fenchurch spec.
• Put a generated image on a spatial board: Canvas.
• Build a page around it: Workbook.