I Lost $1,200 on Etsy Returns Because My Images Were Too Small: A Print-on-Demand Resolution Guide

October of last year, I got my first Etsy return request. A woman in Oregon had ordered an 18x24 inch canvas print of a fantasy landscape I generated with Midjourney. She wrote that it arrived looking "like someone stretched a thumbnail across a poster." She was right. I had uploaded the 1024x1024 Midjourney output directly to my print provider without touching the resolution. The print looked awful. I refunded her $42 and felt sick.

That was return number one of seventeen. Over the next three months, I refunded or replaced $1,247 worth of orders. Every single one was the same problem. I was selling prints at sizes that my source images could not physically support. I had no idea what DPI actually meant in practice. I assumed my print provider would handle the technical stuff. They did not.

By January, I had figured out the math and built a workflow that actually works. I have not had a single resolution-related return since. Here is everything I wish someone had told me before I listed my first product.

The math that tripped me up for months

Here is the only formula you need. Ready? Print size in inches times 300 equals the pixel dimensions you need. That is it. An 8x10 print needs 2400x3000 pixels. An 18x24 poster needs 5400x7200 pixels. A 4x4 coaster needs 1200x1200 pixels. If your image has fewer pixels than this, the print provider will stretch each pixel across multiple ink dots on the paper. That is what makes prints look blurry.

Most AI image generators output at 1024x1024 or maybe 2048x2048 if you are using the latest versions. At 300 DPI, a 1024x1024 image prints at 3.4 inches. That is smaller than a postcard. If you list it as a 12x18 poster, you are stretching each pixel to cover more than five times the area it was designed for. The math simply does not work.

I spent two months blaming my print provider, switching between Printful and Printify and Gooten, thinking one of them would magically fix my low resolution files. They all produced the same blurry results because the problem was never the printer. It was the file I was giving them.

What AI upscaling actually does to your pixels

When you upscale an image with AI, the model does not just stretch what is already there. It looks at your 1024x1024 image and predicts what it should look like at 4096x4096. It fills in the gaps between your existing pixels with new pixels that make sense based on what it learned from millions of real photographs.

The important thing here is that different upscalers fill in those gaps differently. Basic bicubic resizing in Photoshop just averages neighboring pixels together. You get a bigger image but it looks soft and smeary. Real-ESRGAN and similar AI models actually generate new texture detail. A wool sweater still looks like wool at 4x. Brick walls keep their grain. Wood grain does not turn into brown plastic.

For print-on-demand, this texture preservation is everything. Most POD products are things people touch. Pillows, blankets, canvases, t-shirts. If the texture is gone, the product feels cheap even if the design is good. I learned this the hard way with a fleece blanket order where a forest scene looked like green and brown blobs because the upscaler I used had smoothed all the tree bark texture into nothing.

The exact sizes you need for every product type

After testing dozens of products across three platforms, here are the pixel dimensions that actually work. Not the minimums the platforms list on their help pages. The dimensions where customers stop opening return requests.

For framed prints and canvases, 300 DPI is non-negotiable. An 8x10 needs 2400x3000. An 11x14 needs 3300x4200. A 16x20 needs 4800x6000. For 18x24, which is the most common poster size on Etsy, you need 5400x7200. That is a lot of pixels. A 1024x1024 Midjourney output upscaled 4x only gets you to 4096x4096, which covers up to about 13 inches at 300 DPI. For anything larger, you may need to upscale again or accept 200-250 DPI, which still looks decent on a poster viewed from a few feet away.

For apparel, the requirements are different. A standard t-shirt print area is roughly 12x16 inches. That means 3600x4800 pixels at 300 DPI. But here is the thing about apparel. Screen printing and direct-to-garment printing are more forgiving than paper prints. The fabric texture naturally hides some softness. I have gotten away with 200 DPI on t-shirts multiple times and the results looked fine. Do not push it below 200 though. Below 200 DPI on a shirt, you start seeing jagged edges on text and thin lines.

For all-over-print products, things get intense. An all-over-print hoodie might need a 4000x5000 pixel file at minimum. Some providers ask for 12000 pixels on the long edge for large tapestries. Those sizes are not achievable from a single 1024px AI generation even with 4x upscaling. For those products, you need to generate at the highest base resolution your AI tool allows, upscale as much as quality allows, and then honestly evaluate whether the result is sharp enough. If you see any softness at 100 percent zoom, it will be visible on the product.

For small items like stickers, magnets, and coasters, you can relax a little. A 3x3 sticker at 300 DPI only needs 900x900 pixels. Most AI outputs can handle that without any upscaling at all. The problems start when you try to stretch a 1024px image across an 18 inch poster.

The workflow I use for every listing now

I have boiled it down to four steps because every extra step is another chance for me to mess something up. I have messed up a lot of things. I try to keep it simple now.

First, I generate at the highest resolution the AI tool allows. Right now that means using Midjourney's highest quality settings or Stable Diffusion with high-res fix enabled. I aim for at least 1664x1088 or ideally 2048x2048 as a starting point. Every extra pixel in the source image means fewer pixels the upscaler has to invent from scratch.

Second, I upscale with ClarifyPix AI upscaling at 4x. For a 2048x2048 source, that gives me 8192x8192. That covers almost every print size I sell up to 27 inches at 300 DPI. Cost is 4 credits per image. Processing takes about 8 seconds. This is the step that took me from 17 returns to zero.

Third, I check the upscaled image at 100 percent zoom before uploading. I look at the edges. I look at any text. I look at faces if there are any. If faces look soft, I run a second pass through face restoration. That adds 4 more credits and 5 more seconds. Worth it for portraits. Skip it for landscapes.

Fourth, I download as PNG and upload directly to my print provider. Never JPEG at this stage. JPEG compression introduces artifacts at exactly the wrong moment. You spend credits getting a clean upscale and then JPEG compression undoes some of that work before it reaches the printer. PNG keeps every pixel you paid for.

What I got wrong about mockups

For the first three months, I used the same low resolution files for my Etsy listing mockups. The mockups looked fine on screen because screens only show 72 to 150 DPI. A blurry print file can look perfectly sharp on a phone screen. That is why I did not catch the problem before customers did.

Now I upscale before making mockups too. The mockup tool I use places my design onto a photo of a framed print or a canvas. If the design file is sharp, the mockup looks realistic. If the design file is soft, the mockup still looks fine on a screen, and I do not know there is a problem until the customer emails me. So I upscale first, make mockups second, and I check every mockup at 100 percent zoom on a desktop monitor before I list anything. A phone screen will lie to you. A desktop monitor at full zoom usually tells the truth.

I also stopped using the same upscaled file for every product variant. A 16x20 canvas and a 5x7 greeting card should probably come from different upscales, or at least the greeting card version can be a downscaled copy of the large upscale. Using one massive upscaled file for everything is wasteful on storage but it does guarantee every variant is at least as sharp as it needs to be. I keep one master upscaled file per design and resize down from there for smaller products.

When upscaling cannot save you

AI upscaling is good but it is not magic. If your source image is fundamentally bad, upscaling makes a bigger, sharper version of a bad image. A composition with muddy colors and poor contrast at 1024 pixels will still have muddy colors and poor contrast at 4096 pixels. The resolution goes up but the quality of the art does not.

Also, if your source image has visible AI artifacts, those get sharper and more noticeable when you upscale. Weird fingers become very crisp weird fingers. Melting faces become high-resolution melting faces. Fix all the visible problems in your source image before you upscale. Upscaling is the second step, not the first.

There is also a practical ceiling on how far you can push a single upscale. Going from 1024 to 4096 with a 4x upscale usually works great. Going from 1024 to 8192 with an 8x upscale is pushing it. The AI has to invent more and more pixels with less and less original information to guide it. At some point, around 6x to 8x depending on the image, the invented detail starts to look artificial. You get textures that repeat too perfectly or smooth areas that look airbrushed. For large format prints, I stick to 4x upscaling and if I need bigger, I generate a larger source image rather than stacking multiple upscales.

If you are selling prints, canvases, shirts, or anything else where a real person will hold your image in their hands, resolution is not optional. It is the difference between a product someone keeps and a product they return. I learned that the expensive way. Do not be me. Upscale your files before you list them. Your Etsy rating will thank you.