The Fold · From the factory floor
How to take a photo of packaging from ChatGPT and put it on a real dieline
The picture in your chat window is the easy part. Turning it into a file a press can cut is the part everybody gets stuck on — and it takes about fifteen minutes.
· Nick Wahl, founder
You described a package to ChatGPT. It gave you something better than you expected — the right colours, the right mood, type in the right places. You saved the image, took it to someone who prints packaging, and they asked for your dieline. That is where it stops for most people.
Here is what is actually going on, and how to get from that image to a package on a shelf.
Why the image can’t be printed
An AI packaging image is a photograph of a package that does not exist. It is a flat grid of pixels. Everything a factory needs in order to make something is missing from it — not hidden, not low quality, genuinely not there.
This is not a knock on the image. It is a brief — a very good one. The job is to rebuild what it shows as a real construction at a real size, and that is a different kind of work than drawing a picture.
The five steps
1. Choose the construction and the size first
Before the image matters at all, decide what you are actually ordering: a folding carton, a stand-up pouch, a corrugated shipper, a label. Then set the size — pick a stock size or type your own dimensions.
Do this first because the size is what makes everything else real. It decides the flat blank, how the panels lay out, where the creases fall, and whether the proportions can be folded and glued at all. A design has nothing to sit on until the dieline underneath it exists.
2. Bring the image in
Upload it in the design chat. Two rules, and both come from what the rebuild actually needs:
- One package per photo. A product lineup, a shelf shot or a collage has nothing single to rebuild. If four boxes are in frame, there is no way to know which one is yours.
- Two photos if you want both sides. One front, one back — they go through together as a single conversion, and the back panel is designed to match the front.
3. Convert it into a reference
This is the step nobody expects and the one that makes the rest work. Your image gets rebuilt as your exact package at your exact size — the same design, the same wording, the same colours, now in the shape you are actually ordering.
The reason is simple: ChatGPT rendered whatever proportions looked good in the frame. That is almost never the package you chose. Mapping a tall, narrow render onto a wide, shallow carton stretches the art and pulls the type apart. Converting first means there is something true to map.
Converting is a rebuild, not a reimagining. It reproduces what you gave it on your package. If you hand it a matte black box with a gold monogram, you get a matte black box with a gold monogram — at your size, on your construction.
4. Say how the reference should be used
Two questions decide everything: should the design match the reference or just borrow its feel, and should the text and logos be placed for you or left blank for your own files? Those answers pick one of five ways to run the design.
If you are unsure, take the recreate-the-look route and add your own branding afterwards. Your logo lands pixel-perfect from your own file, your brand name cannot be misspelled, and the design still carries everything you liked about the reference.
5. Design it, finish it, order it
The design round generates real artwork onto your real dieline — not a picture of a package, the flat blank with the art in position, bleed past every knife line and a safe area held clear of every fold.
Then you finish it on the canvas: your exact logo file, your real barcode, your QR code, your nutrition or facts panel, your ingredients and legal copy, placed by you at exact size. None of that is ever generated — a made-up barcode does not scan, and invented nutrition figures are not yours. Check the package in 3D, and order it made.
Getting a better image out of ChatGPT in the first place
Since the image becomes your reference, a few habits make the rebuild land much closer:
- Ask for one package, straight on, on a plain background. A three-quarter hero shot with props hides two panels and adds things that have to be stripped out.
- Name the construction. “A stand-up pouch” or “a straight tuck end carton” renders far closer to what you are ordering than “packaging”.
- Say the proportions. Tall and narrow, wide and shallow, square. It will not be exact, and it does not need to be — closer proportions convert more faithfully.
- Keep the wording short. Image models letter beautifully in short bursts and badly in paragraphs. Brand name, product name, one line of flavour.
- Ask for the back panel separately. One image for the front, one for the back, same brief — then bring both.
What to expect, honestly
Recreating a design faithfully including its lettering is the hardest thing in this whole flow. It can take a rerun or two to land, and it gets shakier on a package that is nearly as short as it is tall. That is why the standard route — everything rebuilt except the text — is the one worth defaulting to. It produces a cleaner file, and you end up with sharper branding anyway.
The other thing worth knowing: small imperfections in a generated design are normal and fixable. Judge a concept on the overall look and direction, then refine the one you like rather than starting over.
Bring your ChatGPT image and see it on a real dieline
Pick your package and size, upload the image, and watch it rebuilt as your exact construction. Your first reference image and your first design round are on us.
Start with your image →Common questions
Can I print a packaging design that ChatGPT generated?
Not directly. What ChatGPT produces is an image — a flat raster with no vector paths, no scale, no fold lines, no glue flaps and no bleed. A printer cannot build a die from it. The image is genuinely useful as a reference for what you want, but it has to be rebuilt as a production dieline before anything can be cut. TWOFORM does that rebuild: it converts your image into your exact package at your exact size, then designs it onto a true dieline you can order from.
Why does the image have to be converted first?
Because the picture is not your package. ChatGPT renders whatever proportions looked good in the frame, which is almost never the shape or the size you are actually ordering. Converting rebuilds that design as your chosen construction at your dimensions, keeping the look and the wording, so there is something true to map onto the dieline. Designing straight from an unconverted photo maps art onto a package that does not match it.
How many photos can I upload?
One package per photo. A single image must show one box, one pouch or one label — a product lineup, a shelf shot or a collage has nothing single to rebuild. You can supply two photos when you want both sides: one front and one back, converted together in a single pass.
Will it keep the text from my ChatGPT image?
That is your choice, and it is the most important decision in the whole flow. The standard route recreates everything except the text and logos, so those areas come back clean and you drop your own files on afterwards — sharper, and nothing can be misspelled. The full-control route letters the package for you and reproduces your logo exactly, which gets you closer to finished in one pass but is harder and can take a rerun or two.
Do I need to know my package size first?
Yes, and it is quick. You choose the construction and either pick a stock size or type your own dimensions, and the dieline is drawn to those measurements. The size is what makes the whole thing real — it decides the blank, the panel layout, the bleed and whether the proportions can be folded and glued at all.
What does it cost to try?
Every account gets one free reference image and one free design round. The free reference covers either path — generating one from your description, or converting an image you already have — and the free design round is separate. After those, references and rounds run on credits or a monthly plan.
Where do the barcode and legal copy go?
You place them yourself on the design canvas, after the design lands. Your real barcode, QR code, nutrition or facts panel, ingredients and legal lines are added there from your own files, at exact size and position. They are never generated — an invented barcode does not scan and invented nutrition figures are not yours.