How to Upscale AI Images for Print (Without Losing Quality)

To upscale AI images for print in 2026: (1) calculate the target pixel size from your print size × 300 DPI. (2) Upscale with Crystal (Replicate) for AI art or Topaz Gigapixel for photos. (3) Convert sRGB to CMYK if your printer requires it. (4) Save as TIFF (uncompressed) or high-quality JPEG. A 16×24" print at 300 DPI needs a 4800×7200 image.

That is the four-step recipe. The rest of this guide explains why each step matters, what changes when your print job is a fine-art canvas versus a tradeshow poster versus a billboard, and which upscaler to reach for depending on whether your source is AI art, photoreal AI output, or a restored old photo. The voice is print-tradesman, not "AI enthusiast" — these are the practical numbers you need to deliver real client work without an embarrassing pixelated print landing on someone's desk.

What DPI do you need for print?

DPI stands for dots per inch, and it is the single number that decides whether your print looks sharp or soft. Three benchmarks cover almost every print job a working creator will encounter in 2026.

• 300 DPI is the industry standard for sharp, held-in-hand print — fine-art prints, photo books, magazine pages, postcards, brochures, business cards, photo album prints. If your work will be viewed from arm's length, target 300 DPI.
• 150 DPI is acceptable for large prints viewed from a distance — posters, gallery banners, trade-show pop-ups, retail signage. The eye cannot resolve more detail than that at 6+ feet, so the extra pixels would be wasted file size.
• 72 DPI is screen-only. Sending a 72 DPI file to a printer will produce visibly pixelated output at any meaningful physical size. This is the "I downloaded the small JPEG and now my print looks terrible" failure mode.

For very large outdoor signage — billboards, building wraps, bus-stop posters — 100 DPI is common because the typical viewing distance is 20+ feet. Most outdoor print houses will accept files at 100 DPI scaled to final size without complaint.

How big can you print at 72 DPI? (the math behind print quality)

The math is straightforward. To find how big you can print a given image at acceptable quality, divide pixel dimensions by your target DPI. A 1024×1024 raw AI image divided by 300 DPI = 3.41×3.41 inches. That is barely larger than a sticky note. The same image at 150 DPI prints at 6.83×6.83 inches, and at 72 DPI it stretches to 14.2×14.2 inches — but it will look soft and pixelated because 72 DPI is below the threshold human eyes resolve detail at normal viewing distance.

This is why every AI image headed for print needs upscaling. A native 1024×1024 generation from Nano Banana Pro, Seedream 4.5, or Midjourney can only print sharply at thumbnail size. For anything larger than a postcard, you have to grow the pixel count before sending the file to the printer.

Step 1: Calculate your target pixel size

Open a calculator. Multiply your print dimensions (in inches) by 300 (or 150 for large posters). Note both numbers — width and height. That is your target.

If you are printing at non-US dimensions, the math is the same — multiply centimeters by your DPI ÷ 2.54 to convert. A 50×70 cm gallery poster at 300 DPI = 5906×8268 pixels. A0 (84.1×118.9 cm) at 150 DPI = 4967×7022 pixels.

One subtlety: if the print is full-bleed (the image extends to the edge of the paper with no white border), add 0.125 inches of bleed on each edge before the math. A 16×24" full-bleed print is actually 16.25×24.25" of working canvas, which at 300 DPI = 4875×7275 pixels. Most print houses will reject jobs without proper bleed because the trimmer cannot cut a perfectly aligned edge — bleed gives the cutter room to be off by a millimeter without exposing white paper.

Step 2: Pick the right AI upscaler for the source type

Not every upscaler is built for every source. The wrong one will smooth away the texture that made your AI image interesting in the first place, or hallucinate fake detail that breaks photoreal credibility. Match the tool to the source:

• AI art and illustrations → Crystal (Replicate) or Magnific creative. Crystal preserves the painterly textures and brush-mark feel the AI generated. Magnific can push creativity further but may drift the look away from the original. For clean enlargement, Crystal. For artistic embellishment, Magnific.
• Photoreal AI images → Crystal or Topaz Gigapixel. Crystal handles up to 200 megapixels at 2x–10x in a single pass and is the cleanest upscaler for AI-generated source. Topaz Gigapixel adds optional face recovery and detail enhancement that helps for photoreal portraits.
• Old photos and restored work → Topaz Photo AI first, then Crystal or Magnific. Run face recovery and denoising at original resolution before upscaling. Upscaling damaged input amplifies the damage.
• Vector logos and brand marks → no upscaling needed. If the source is already vector (Recraft V3, Adobe Firefly Text to Vector, hand-drawn SVG), it scales infinitely without loss. Re-export at the print size from Illustrator or Affinity Designer.

Crystal (Replicate) is the cleanest upscaler for AI-generated source images — it doesn't smooth over the textures the AI added. Rangy gives you Crystal through your own Replicate API key at ~$0.025 per upscale. For face-heavy print work, you'll still want Topaz Photo AI on your desktop — they're complementary, not interchangeable.

For a deep comparison of the three leading upscalers side by side, read Topaz vs Magnific vs Crystal. For the broader market, see our roundup of the best AI image upscalers 2026.

Step 3: Convert sRGB to CMYK (if your printer needs it)

Every AI image generator outputs in sRGB — the standard RGB color space designed for screens. Print uses CMYK — the four-ink color space of cyan, magenta, yellow, and black. CMYK has a smaller color gamut than sRGB, which means some vivid screen colors (especially saturated greens, blues, and purples) simply cannot be reproduced in print. They will shift toward the nearest reproducible CMYK approximation.

When you do not need to convert: consumer print services like Mpix, Bay Photo, Costco Photo, Whitewall, Snapfish, Walmart Photo, and most online photo labs accept sRGB and handle the conversion internally with their lab printer's profile. Inkjet prints from your own Epson or Canon photo printer also work directly from sRGB.

When you do need to convert: professional offset print houses for magazines, books, brochures, packaging, and high-end art prints will demand CMYK files matched to a specific paper-stock ICC profile (commonly GRACoL 2013 for coated US stocks, ISO Coated v2 for European). Convert in Photoshop via Image → Mode → CMYK Color, or in Affinity Photo via Document → Convert Format, then assign the ICC profile your printer supplied.

Always do the CMYK conversion after upscaling, not before. Upscalers expect sRGB input and produce sRGB output. Converting first and then upscaling will damage color accuracy on the second pass.

Step 4: Export the right file format (TIFF vs JPEG vs PDF)

The file format you save matters as much as the upscaling. Three formats cover every print job:

• TIFF (uncompressed) — the lossless archival format. Files at 16×24" 300 DPI run 100–500 MB depending on bit depth. Use for fine-art prints, gallery work, museum reproductions, and anything you might re-print at a larger size later. Supports 16-bit color depth, which preserves smooth gradients and protects against banding in skies and soft skin tones.
• High-quality JPEG (quality 90+) — the everyday print format. Compresses to roughly 10–30 MB for the same image. Visually indistinguishable from TIFF at quality 95+, dramatically smaller file. Use for everyday photo prints, social posters, business deliverables, and anything where storage or upload bandwidth matters.
• PDF/X-1a — the professional offset print format. Embeds CMYK profiles, fonts, and the image in a single self-contained file the print house can run without missing assets. Required by most commercial print shops. Export from InDesign, Illustrator, or Affinity Publisher.

The single biggest export mistake is re-saving JPEGs multiple times. Each JPEG save re-compresses the image, and the artifacts compound. If you generate as PNG, upscale to PNG, then convert to JPEG, the file is fine. If you generate as JPEG, edit in Photoshop, save as JPEG, upscale, save as JPEG, then send to the printer — you have stacked four compression passes and the print will show it.

Common print mistakes — and how to avoid them

Four print mistakes show up over and over again when AI creators send files to commercial printers. Each one is fixable in two minutes if you catch it before submission.

1. Ordering a large print with screen-resolution input. A 1024×1024 AI image at 300 DPI prints sharply at 3.4×3.4 inches. Ordering a 16×24" print from a file that small is the most common failure. Always upscale first to the calculated target.
2. Forgetting CMYK conversion for offset jobs. Submitting an sRGB file to an offset press leaves the printer guessing your intent — the result will lose saturation in vivid colors. If your job ticket says CMYK, deliver CMYK.
3. Re-saving JPEGs multiple times. Each save compounds compression artifacts. Generate as PNG, upscale to PNG or TIFF, convert to JPEG once at the end.
4. Not adding bleed for full-bleed prints. If your image is supposed to extend to the paper edge, add 0.125 inches of overflow on every side. Without bleed, the trimmer will cut white paper into your final piece.

One additional warning specifically for Magnific users: the creativity slider can introduce color drift. Setting creativity above the default risks the upscaler inventing detail that does not match the source — and worse, shifting overall color balance toward whatever the model thinks "looks better." Always verify with a small test print before mass production. For more general technique, see how to upscale images without losing quality.

The verdict — your print prep workflow

The most common AI print job a working creator will field in 2026 is a poster from a 1024×1024 generation. Here is the verdict workflow for that exact case.

If the source is photoreal and faces matter, swap Crystal for Topaz Gigapixel with face recovery on the upscale pass — and accept the file goes from $0.025 to roughly $0.30 in Topaz subscription cost amortization, plus a slightly different texture feel. If the source is a restored old photo, run Topaz Photo AI first on the original to recover faces and reduce noise, then Crystal for the final upscale. For background reading on the restoration half of the pipeline, see our best AI photo restoration tools guide.

The fundamentals do not change much from job to job: get to the right pixel count with the right tool, match the color profile to the printer, save the right file format, add bleed, order a proof. Once the recipe is in your head, every job is a 10-minute exercise rather than a guessing game. For the broader landscape of which AI image generator to use as your source, see the complete 2026 AI image generator guide.

Frequently asked questions

Last updated May 2026. Color profiles and printer requirements vary — always print a small test before committing to a large run.