TLDR
For a standard MTG card, your image needs enough pixels. As a baseline, 750 × 1050 px covers trim at 300 PPI. If your source is 480 px tall, it’s not “low-res,” it’s a rumor.
Upscale first, sharpen last. Doing it in the opposite order is how you get that “deep-fried” crunchy look.
Avoid repeated JPEG saves. That’s how you get mosquito noise, blocky gradients, and sadness.
Gradients need a 16-bit workflow (or added dithering/noise) or you get banding, especially in skies.
Do a tiny test print (even just one card) before you commit to printing a whole Commander deck that looks like it was exported in 2006.
You can blame your printer all you want. It’s not going to magically discover detail that isn’t in your file. Printers are good, but they’re not archaeologists.
This post is about MTG proxy image prep: how to upscale proxy art without AI artifacts, sharpen it without halos, and keep gradients from turning into a stack of visible stripes.
The “pixels per inch” reality check (aka: your art has to exist)
Let’s get the most common misunderstanding out of the way.
PPI is math, not vibes
If you’re printing a standard Magic card, you’re aiming at 2.5" × 3.5" (trim). The pixels you need depend on the print resolution you’re targeting, but 300 PPI is the common baseline for high-quality print work.
So at 300 PPI:
2.5" × 300 = 750 px
3.5" × 300 = 1050 px
That means:
750 × 1050 px is “baseline acceptable” for a full card face at trim.
If you’re adding bleed, you need more pixels (and yes, bleed matters if you want to avoid white edges after cutting, but that’s a separate rabbit hole).
Now notice what isn’t on that list:
“72 DPI, but I changed it to 300 so it should be fine.”
Changing the DPI/PPI number in an image editor without adding pixels is like changing your character’s name and expecting their stats to improve.
Quick rule of thumb
If the smallest dimension is under ~1000 px, you’re almost always going to need either (1) a better source, or (2) a careful upscale workflow.
If the art includes text, low-res becomes obvious fast. Text is a snitch.
Upscaling proxy art: what works, what artifacts to avoid
Upscaling is the process of making an image larger while keeping it looking clean. There are two broad approaches:
Traditional resampling (bicubic, Lanczos, etc.)
AI super resolution (Adobe, Real-ESRGAN, Topaz, etc.)
Here’s the honest truth: AI upscaling can look great, and it can also invent nonsense. Your job is to use it like a tool, not a lottery ticket.
Good / Better / Best framework for upscaling
Good: Find a better source
If you can re-export the original art at a higher resolution, do that.
If you’re using custom art you own, export again from the original file (not the screenshot of the preview of the JPEG you saved three edits ago).
Better: Traditional upscale (when the art is already clean)
Works well for clean digital art, flat colors, and vector-like edges.
Tends to look softer, but also tends to avoid “AI glitter texture.”
Best: AI super resolution (when you’re stuck with small art)
Best for tiny sources or art that needs edge reconstruction.
Also the most likely to create weird artifacts if you push it too hard.
Common upscaling artifacts (and what they look like on cards)
If your proxy art looks “crunchy,” it’s usually one of these:
Ringing halos: bright outlines around edges, especially around black text or hard borders.
Waxy smoothing: faces, clouds, and soft shading look like they got sanded down.
Hallucinated texture: “detail” appears that wasn’t in the original. Sometimes it looks cool. Sometimes it looks like your dragon has psoriasis.
Block compression ghosts: JPEG noise gets “enhanced” into visible patterns, especially in skies and gradients.
Practical upscaling workflow (that avoids most pain)
Clean first
If the source is a JPEG with heavy compression, run a light noise reduction or artifact cleanup before upscaling. Otherwise, the upscaler will lovingly preserve every blocky mistake.
Upscale in 2× steps
Going straight to 6× is how you get fantasy texture soup. Do 2×, check it, then 2× again only if needed.
Pick the right model
“Photo” models can do strange things to illustrated art.
“Art/CG” models can preserve edges better, but may posterize gradients if pushed.
Inspect at 100%
Always judge sharpening and artifacts at 100%. Zoomed out hides sins.
Stop when it looks real
Don’t chase microscopic detail on a card that will be viewed at arm’s length in a sleeve under warm kitchen lighting.
Sharpening for print: crisp proxies without haloing
Sharpening is not “adding detail.” Sharpening is increasing edge contrast so the image appears clearer.
That’s why oversharpening looks crunchy. You didn’t make it more detailed. You just made the edges louder.
The order matters
Resize and upscale first
Then sharpen
Then export
If you sharpen first and then upscale, you amplify the halos and noise. It’s the visual equivalent of yelling into a microphone and then turning the speakers up.
The two sharpening mistakes that wreck proxies
Mistake #1: Sharpening the whole image
You usually only need sharpening on edges and fine detail, not on smooth gradients.
Sharpening smooth areas creates grain and banding.
Mistake #2: Using too much radius
For card-size images, the radius needs to be small. Big radius sharpening creates visible halos that scream “I edited this.”
A simple, repeatable sharpening recipe
This is a starting point, not a religion:
Use Smart Sharpen or Unsharp Mask
Keep radius low
Think 0.3 to 0.8 px radius as a starting range for card-face images.
Increase amount slowly
You can push amount, but keep radius tight to avoid halos.
Avoid sharpening noise
Use threshold (Unsharp Mask) or mask the sharpening layer so it hits edges and text, not flat color fields.
If colors get crunchy
Apply sharpening to luminosity only (so you don’t get color fringing on edges).
If your card frame text is part of the image and not actual text, sharpening can help readability. But if the underlying text is truly low-res, the best fix is often re-rendering the text at a higher resolution, not trying to “enhance” it into legibility.
“Why does my art look crunchy?” The usual culprits
Crunchy art is almost always cumulative damage. One bad step is survivable. Three bad steps is a lifestyle.
Culprit list
You upscaled and then sharpened twice
Many AI upscalers add their own sharpening. Then you add sharpening. Then you add “Clarity” because it “pops.” Now everything has edges like broken glass.
You resized down with a method that adds sharpening
Some downsampling settings bake in extra edge contrast. Great when intended, brutal when accidental.
You edited in 8-bit and pushed contrast
Gradients break. Shadows posterize. You get banding and crunchy transitions.
You saved JPEG multiple times
Each save adds compression artifacts. Upscalers turn those artifacts into “detail.” They are wrong.
The “undo the crunch” fix
If you already have a crunchy file:
Go back to the earliest, least-processed version you have.
Do artifact cleanup (light noise reduction if needed).
Upscale.
Sharpen lightly at the final size.
Export once.
If you don’t have an earlier version, you can still salvage it, but you’re basically doing restoration work. Fun for exactly nobody.
Avoiding banding and posterization in gradients (why the sky looks broken)
Banding happens when a smooth gradient gets quantized into visible steps. Skies, fog, soft shadows, and “nice subtle glow” effects are prime victims.
Why it happens
8-bit color has limited steps per channel
JPEG compression makes gradients worse
Heavy editing (especially curves and contrast) stretches those steps until you can see them
And yes, banding can show up more in print because paper doesn’t hide sins the way a backlit screen does.
Fixes that actually work
Work in 16-bit while editing
Do your gradient edits in 16-bit mode, convert down only at the end if you must.
Add subtle noise or dithering
A tiny amount of noise breaks up the visible steps. It feels wrong, but it prints right.
Avoid heavy JPEG compression
If a gradient is important, avoid turning it into a JPEG punching bag.
A quick “sky salvage” trick
If you have a banded sky gradient:
Duplicate the layer.
Apply a very small noise (monochromatic noise often works best).
Lower opacity until the banding visually disappears.
You’re not trying to make it grainy. You’re trying to make it continuous.
The 10-minute proxy image preflight checklist
Before you print a whole deck, do this once. Future-you will be annoying about how grateful they are.
Resolution check: is the art at least 750 × 1050 px for card trim (baseline)?
Compression check: does it have obvious JPEG blocks, mosquito noise, or crunchy edges?
Upscale check: if upscaled, did you inspect at 100% for hallucinated texture and halos?
Sharpen check: is sharpening subtle, edge-focused, and done at final size?
Gradient check: any skies, fog, soft shadows. If yes, did you prevent banding (16-bit workflow or dithering/noise)?
Export check: did you export once, from the master file, without repeated JPEG saves?
Reality check: print one test card (or even a cropped strip) before committing to 100 cards.
If you don’t want to babysit files
If you’re ordering proxies from PrintMTG using a decklist, the heavy lifting is mostly handled for you. You’re not uploading a folder of questionable images, you’re telling us what you want printed.
If you’re designing custom cards with uploaded art, then this article is your friend, because “garbage in, garbage out” is not just a saying, it’s the entire printing industry’s personality.
Also, quick reminder: proxies are for casual play and playtesting. Sanctioned events are a different universe with different rules.
FAQs
What resolution do I need for MTG proxy cards?
As a baseline, 750 × 1050 px covers a standard card face at 300 PPI (2.5" × 3.5"). If you’re adding bleed, you need more pixels. If your source is smaller than this, expect softness unless you upscale or find a better source.
Does changing DPI/PPI in Photoshop make my proxy higher quality?
Not by itself. If you didn’t add pixels, you didn’t add detail. The pixel dimensions matter.
Should I AI-upscale rules text and mana symbols?
If the text is baked into a low-res image, AI can help a little, but it can also invent shapes and create artifacts. If you can re-render the text cleanly (or rebuild the card in a proper template), that usually looks better.
Why do my gradients band even if I’m at 300 PPI?
Banding is usually a bit depth and compression problem, not a resolution problem. Work in 16-bit while editing gradients and add subtle dithering/noise if needed.
Why do my proxies look sharp on screen but not in print?
Screens are backlit and forgiving. Print is honest and frequently rude. Oversharpening can also look “sharp” on screen but turns into halos and crunchy edges in print. Test print a sample.