AI vinyl sleeve art at 12 inches: gpt-image-2.5 stops at 240 dpi

A 12-inch square sleeve face at 300 dpi needs 3,600x3,600 pixels, which is 12,960,000 pixels. openai/gpt-image-2.5 caps a request at 8,294,400 pixels, so its largest square is 2880x2880: 240 dpi on a 12-inch face. To reach 300 dpi you either upscale a generated image or choose a model tier that renders larger.

Check the 12-inch figure against your pressing plant's file guide; this page only does the pixel arithmetic.

The arithmetic (read 2026-10-04)

The Sume docs and OpenAI's image guide give the same custom-size rule: edges in multiples of 16, maximum edge 3840, aspect ratio at most 3:1, 655,360 to 8,294,400 total pixels. A square hits the pixel cap before the edge cap.

Option 1: a model with a 4K tier

Sume's Image API lists a resolution tier of 512, 1K, 2K or 4K, and the Nano Banana models carry the 0.5K to 4K tiers. Google's pricing page describes the Nano Banana Pro 4K tier as up to 4096x4096 pixels at $0.24 per image on Google's own API, against $0.134 for 1K and 2K. 4096 pixels over 12 inches is about 341 dpi. Confirm the delivered pixel size on a test job before you rely on it, and read Sume's cost_usd pricing line for what Sume charges.

Option 2: generate, then upscale

Generate at 2880x2880, pick the best take, and run a separate upscale step. Sume's hosted tool list names image_upscale_create; the Image API itself does not take an upscale factor, and size accepts a resolution tier rather than custom pixels.

side_in = 12
for px in (2880, 3600, 4096):
    print(px, round(px / side_in), "dpi")

Which to choose

If the art has fine linework or small type, a native 4K render avoids upscaling artefacts on the details. If it is painterly or photographic texture, 240 dpi plus a clean upscale is usually the cheaper route. Whichever you pick, put the title and track list on the sleeve in code at full resolution rather than asking the model to draw them.