The art is the easy part now. You describe a scene, wait forty seconds, and get something you would have paid an illustrator three figures for two years ago.
Then you try to sell it as an A1 poster and the whole thing falls over — not because the image is bad, but because there simply are not enough pixels in it. This is the part nobody mentions in the "make £5k a month with AI art" videos, and it is the only part that actually decides whether a customer keeps the print or returns it.
What Midjourney actually hands you
Midjourney targets roughly one megapixel at the base render. That figure is spread across whatever aspect ratio you asked for, so the exact dimensions move but the total does not:
| Aspect ratio | Base render | After 2× upscale | After 4× upscale |
|---|---|---|---|
| 1:1 (square) | 1024 × 1024 | 2048 × 2048 | 4096 × 4096 |
| 2:3 (portrait) | ~832 × 1248 | ~1664 × 2496 | ~3328 × 4992 |
| 3:2 (landscape) | ~1248 × 832 | ~2496 × 1664 | ~4992 × 3328 |
One megapixel sounds like plenty until you convert it into inches. Pixels only become a physical size when you fix a DPI, and 300 DPI is the print-industry default for anything viewed close up.
| Midjourney output (2:3) | Biggest sharp print at 300 DPI | Biggest acceptable print at 150 DPI |
|---|---|---|
| Base, ~832 × 1248 | 2.8 × 4.2 in — smaller than a postcard | 5.5 × 8.3 in — roughly A5 |
| 2× upscale, ~1664 × 2496 | 5.5 × 8.3 in — roughly A5 | 11.1 × 16.6 in — roughly A3 |
| 4× upscale, ~3328 × 4992 | 11.1 × 16.6 in — roughly A3 | 22.2 × 33.3 in — roughly A1 |
So the honest ceiling is this: Midjourney on its own gets you to about A3 at full quality, or about A1 if you accept 150 DPI. A0 is out of reach without a separate upscaling step. If you want the reasoning behind why 150 DPI is genuinely fine on a large poster but not on a greetings card, the DPI and resolution guide works through it.
You can check any generation against every poster size in one pass with the free print resolution checker — it reads the dimensions in your browser and tells you which sizes you can safely enable.
The aspect ratio trap
This one costs more sales than resolution does, because it is silent.
Generate at --ar 2:3, list it as A3, and the printer has to reconcile a 1:1.5 image with a 1:1.414 sheet. Something has to give: either the image is cropped, or it is squashed, or a white band appears. Whichever the printer chooses, it chose it, not you — and the composition you spent an hour prompting gets its edges eaten.
Midjourney only accepts whole-number ratios, so matching print sizes exactly takes a little arithmetic:
| Print size family | True ratio | Midjourney flag | Error |
|---|---|---|---|
| A-series — A5 through A0 | 1:1.4142 | --ar 17:24 | 0.17% |
| A-series, quick version | 1:1.4142 | --ar 5:7 | 1.0% |
| 12×18, 16×24, 24×36 in | 1:1.5 | --ar 2:3 | exact |
| 18×24 in | 1:1.333 | --ar 3:4 | exact |
| 8×10, 16×20 in | 1:1.25 | --ar 4:5 | exact |
| 11×14 in | 1:1.273 | --ar 11:14 | exact |
| Square prints | 1:1 | --ar 1:1 | exact |
The practical consequence: if you want to sell one design in both A-series and US inch sizes, you cannot use one file. The ratios genuinely differ. Either generate twice, or compose with enough dead space at the edges that both crops survive — which is a design decision you have to make before you generate, not after.
Upscale once, and upscale early
The order of operations matters more than the tool.
Enlarging an image is not the reverse of shrinking one. Shrinking discards information you already had; enlarging manufactures information that was never there, by estimating what should sit between the pixels it can see. That estimate is always a blur, because a smooth average is the safest guess. The master file guide sets this out in full, and it applies with double force to AI art, because upscalers trained to add plausible detail will happily add detail that was not in the original.
Three rules follow:
- Upscale the raw generation, not the finished design. Once you have added text, borders and a signature, an upscaler has to invent those too — and it will round off the letterforms.
- Upscale once. Two passes through different upscalers compound each one's guesses. The tell is a faint bright rim around high-contrast shapes, which is the model over-sharpening its own invention.
- Then build at full size. Do the composition work at the final pixel dimensions, so every element you add is genuinely sharp rather than enlarged along with everything else.
If you would rather keep the whole loop in one place, PSDmate's AI art generator for print runs FLUX, fal.ai, Gemini, GPT Image, Stability and Replicate on your own provider key, with upscaling and a pre-publish resolution check against the sizes you have actually enabled. Midjourney is not one of the six — it has no public API — so a Midjourney workflow means exporting from Discord and bringing the file in, which is fine, it is just a manual step you should plan for.
Where AI art still fails at print size
Resolution is the loud problem. These are the quiet ones, and they only appear on paper:
- Text. Every image model still garbles lettering at small sizes. Set type yourself in Photoshop, Illustrator or Affinity — never prompt for it on a piece you intend to sell.
- Hands, symmetry and repeated motifs. A wonky hand is a joke on a screen and a refund on a wall. Zoom to 100% and look at every region before you commit.
- Near-black shadow detail. Models often crush shadows into flat black, which looks dramatic on a backlit screen and looks like a printing fault on matte paper. Lift the darkest tones slightly before export.
- Colour space. Models output sRGB. Keep it there. Converting to CMYK yourself, then handing that to a printer that converts again, shifts the colour twice — see which file format to export for the rest of that decision.
- Compression artefacts around fine grain. AI-generated texture and JPEG compression interact badly. Export at quality 90–95 rather than the default 80.
The legal bit, briefly
Midjourney's terms give paying subscribers ownership of what they generate, which is what makes commercial POD possible at all. That is Midjourney's position on Midjourney's output. It is not the whole picture.
Prompting a living artist by name, or generating something recognisably derived from a trademarked character, creates takedown exposure regardless of who owns the pixels — and on Etsy a takedown is handled by Etsy, not by a court. Copyright and trademark for POD sellers covers how those reports actually play out and what a repeat strike does to a shop. Etsy also expects AI work to be described accurately in the listing.
None of this is legal advice, and both sets of terms change. Read them yourself before you build a catalogue on them.
The workflow that holds up
Once the art exists, the bottleneck moves. A single wall-art design typically becomes five or six framed mockups, in two or three room scenes, across four or five print sizes — and if the art took forty seconds to make, spending an afternoon on the listing images defeats the point.
- Generate at the correct
--arfor the size family you intend to sell. - Pick the frame you actually like, then upscale it once, to the largest size in the range.
- Compose, set any type, and check at 100% zoom — not fit-to-screen.
- Export sRGB JPEG at quality 90–95, under whatever file-size cap your printer imposes.
- Run it through a resolution check against every size you plan to enable, before you enable them.
- Render the listing images in bulk from your own PSD templates rather than one at a time — the wall art mockup guide covers which scenes convert, and the bulk mockup generator covers doing all of them in one pass.
- Generated at an aspect ratio that matches the print size family, not the default
- Upscaled once, on the raw image, before any design work
- Effective DPI checked at the largest size you plan to enable, not the smallest
- Zoomed to 100% and checked for hands, garbled text, halos and crushed shadows
- Any lettering set by hand, not prompted
- Exported sRGB, JPEG quality 90–95
- Listing describes the work as AI-generated
- No living artist's name and no trademarked character anywhere in the prompt
The short version
Midjourney's constraint is not quality, it is pixel count. One megapixel at base, roughly A3 at 300 DPI after a 4× upscale, roughly A1 if you accept 150.
Match the aspect ratio before you generate, upscale once and early, and check the effective DPI against the biggest size on the listing rather than the one that happens to sell first. The design that sells at A5 all year is the same design that will eventually sell at A1 — and that is the order you will not want to have guessed at.
Frequently asked questions
Can you sell Midjourney images on Etsy?
Midjourney's terms grant paying subscribers ownership of the images they generate, so commercial use on Etsy is permitted under Midjourney's own rules. Etsy's rules are separate: AI-generated work must be listed accurately, and prompting a living artist's name or a trademarked character creates a takedown risk regardless of who owns the output. Check both sets of terms yourself before you build a shop on them.
What resolution does Midjourney output?
The base render targets roughly one megapixel — about 1024 × 1024 for a square, or near 832 × 1248 for a 2:3 portrait. Midjourney's own upscalers multiply that by two or four, so the practical ceiling is around 3300 × 5000 pixels. At 300 DPI that is roughly A3; at 150 DPI it stretches to about A1.
What aspect ratio should I use for A4 and A3 posters?
The whole A-series shares one ratio, 1:1.414. Midjourney only accepts whole numbers, so use --ar 17:24, which is within 0.2% of exact. The commonly suggested --ar 5:7 is 1% out, which costs you a few millimetres of crop at A2 and above.
Do I need a separate upscaler for large posters?
For anything above roughly A2 at 300 DPI, yes. Midjourney's built-in upscale tops out below what A1 and A0 need at full resolution. A dedicated upscaler run once, on the raw generation, before you build the design around it, closes the gap.
Why does my Midjourney poster look soft when printed but sharp on screen?
Because a phone screen shows the file at a fraction of its printed size. Softness in AI art is usually one of two things: the file was enlarged after the design was finished, or the upscaler invented detail that reads as a faint halo around high-contrast edges. Both are invisible at fit-to-screen and obvious at arm's length on paper.