Learn
Make print-ready designs by chatting with ChatGPT or Claude
Most AI tools hand you a picture. A printer needs a file: right trim size, bleed on every edge, CMYK color, 300 dpi, text kept off the cut. Popcorn Editor closes that gap. Connect ChatGPT or Claude to your account once, and from then on you can describe a print job in a chat and get back a real, editable design that is print-correct by construction.
The trick is not a clever prompt. It is giving your assistant a spec, and a chat is a very good place to carry a spec.
What you need
- A paid Popcorn plan. Connecting AI agents is a paid feature.
- An AI client that supports MCP: Claude on web, desktop or mobile, ChatGPT on the web in developer mode, or a terminal or IDE agent like Claude Code, Gemini CLI, Cursor or VS Code.
- About two minutes for the one-time connection.
Not sure which of your apps can do this? See Where you can connect. Wondering what you are handing over? Is it safe to connect an AI agent? answers that first.
Connect once
In Popcorn, open Settings and find Your agent access. It shows your connection URL and, if your client needs one, mints an agent token.
Then follow the guide for your app: Connect Claude, Connect ChatGPT, Connect Gemini CLI, or other clients. You do this once per app, not once per design.
Brief it with a spec, not a vibe
"Make me a flyer" gets you a flyer-shaped picture. A spec gets you a file. Every line you add below maps to something real in the document your assistant writes:
| Say this in your prompt | What it sets up |
|---|---|
| The trim size in mm, for example 85 x 55mm | The finished page size, in print units |
| "3mm bleed" | Artwork extends past the trim by 3mm, which becomes the PDF bleed box |
| "3mm safe margin" | A keep-inside guide, so nothing important sits near the cut |
| "CMYK" | The document works in process ink, not screen color |
| "300 dpi" | Print resolution, instead of the screen default of 72 |
| "Pantone-style spot color for the foil" | A spot swatch, which exports as its own separation plate |
| "Kiss-cut outline for a sticker" | A cut contour drawn on a spot color, the way a printer expects a dieline |
| "A QR code linking to example.com" | A real vector barcode, not a pasted image of one |
Two more that save a round trip:
- Give it the actual words. Real names, real dates, real phone numbers. Placeholder copy produces placeholder layout.
- Say where photos come from. Your assistant can place an image if you give it a URL. If you do not, ask it to leave a described placeholder frame and you drop your own photo in later, in the editor.
A prompt you can paste
Swap the details for your job:
Make me a double-sided business card in Popcorn.
Print spec: 85 x 55mm trim, 3mm bleed, 3mm safe margin, CMYK, 300 dpi.
Front: name "Lena Ortiz", role "Ceramicist", studio name "Kiln & Co".
Back: [email protected], +31 6 12 34 56 78, kilnandco.example,
and a QR code to https://kilnandco.example.
Style: warm, quiet, a lot of white space, one accent color, no gradients.
Before you create it: read the Popcorn design schema and authoring guide,
loop validate-design until it passes, run analyze-design, and render it so
you can check the layout yourself. Then create it and give me the editor link.
That last paragraph is what separates a good result from a first draft. It tells the assistant to use the checking tools rather than guessing.
What your assistant actually does
None of this is magic, and all of it is inspectable:
- Reads the format.
get-schemareturns the Design Document schema, a prose authoring guide, and a set of worked examples, so it is not guessing key names. - Writes a Design Document. A structured description of your card: page size, bleed, color mode, every element and its position in millimeters.
- Validates until clean.
validate-designreturns precise errors. It is free and creates nothing, so the assistant can loop until the document is correct. - Checks the layout.
analyze-designreports things you would normally only catch by eye: elements hanging off the artboard, text overflowing its frame, low-contrast type, elements buried under other elements. - Looks at it.
render-designreturns preview images of each artboard, so the assistant can see its own work and fix what is visibly wrong before you ever open it. - Creates the design.
create-designcompiles the document into a real design in your workspace and returns an editor link, plus a preview of the first page.
Popcorn validates and compiles on the server. The thinking happens in your assistant, using its own model, so this consumes none of your Popcorn AI credits.
Fix it by asking
Open the link, look at it, then go back to the chat:
Tighten the name to one line, drop the accent color to a warmer terracotta,
and move the QR code 4mm further from the trim edge.
Small edits go through a patch, not a rebuild, so the elements you liked keep their identity and your own manual tweaks are not bulldozed. And before every agent write, Popcorn snapshots the previous state as a restorable version called Before agent update, so a revision that goes sideways is one click from being undone.
You export the PDF, not the agent
This is the one manual step, and it is deliberate. There is no export tool on the agent connection. Your assistant builds a document that is print-correct; you open it and produce the file.
Open the editor link, then:
- Run through the preflight checks. Low-resolution images and content outside the safe margin get flagged here.
- Export a print-ready PDF with marks and bleed, choosing PDF/X-4, or PDF/X-1a if your printer asks for it.
Step by step version: How to export a print-ready PDF.
Where this pays off
- Repeat jobs. The same brief with a different name, size or language is a one-line follow-up instead of an afternoon.
- A second format. "Now the same card as an A6 postcard, same palette" reuses everything you already agreed on.
- Starting from a template. Your assistant can search the Popcorn gallery and build on a designed starting point instead of a blank page.
- Unattended checks. A read-only agent token lets a script or an agent inspect and report on designs without being able to change a thing.
Before you send it to print
The assistant is good at the spec you gave it and blind to the one you did not. Worth a look every time:
- Images. Anything you dropped in yourself needs the right resolution. See choosing the right resolution for print.
- Blacks. Large solid areas usually want a rich black, not plain K.
- Bleed on full-bleed art. Backgrounds should run past the trim, not stop at it. See bleed and safe margins.
- Color expectations. Screen color and ink are not the same thing. RGB vs CMYK explains what shifts and why.
Connect once, brief it properly, check the proof, export. Full checklist before you hit send: a pre-flight checklist before you send to the printer. For everything your connected assistant can do, start at the AI agents overview.