Native 4K or 2K image edits: pixel ceilings by model

For the largest edited output on Sume today, GPT Image 2.5 reaches 3840x2160, which is exactly 8,294,400 pixels, its documented ceiling. Ideogram 4.5 stops at its 2K setting. Flux 3 Image, which TechTimes reports as natively producing about 5,456x3,072 (roughly 16.8 megapixels, against 4 MP for Flux 2), is not in the Sume catalog. Pick the model by the pixels you need, then by whether it can edit.

The ceilings side by side

The pixel counts below are arithmetic on figures from the Sume docs and the vendor reports; the Sume column lists only what Sume documents.

The GPT Image 2.5 rules

The Image API docs give the custom-size rules: both edges multiples of 16, the longest edge at most 3840, aspect ratio at most 3:1, and a total between 655,360 and 8,294,400 pixels. 3840x2160 passes all four: 3840 and 2160 are multiples of 16, the ratio is 16:9, and the product is exactly the cap. 4000x3000 fails on the edge limit even though its 12 MP is the size of many photos.

Large and high-quality renders are the ones that most often exceed the 30-second sync wait. The API then returns a 202 job envelope instead of images, so code for 4K should handle both responses. Output tokens are billed at $30 per million, so a bigger canvas costs more than a small one; read the endpoint's pricing record before running a batch.

Do you need native pixels?

A 4K deliverable can come from a native 4K render or from a smaller edit scaled up. The second is cheaper but cannot add detail. If the job is a print, a billboard or a large crop where small text must stay readable, native resolution matters; BFL's example of a soba shop sign with Japanese characters 225 px tall that remain legible at native size is its argument for that. For a web banner, 2K is enough, and Ideogram 4.5's 2K setting at a per-image price independent of size is the cheaper path.

Edits add one more wrinkle. On GPT Image 2.5 and the other models that take references, ask for aspect_ratio: "auto" where the docs allow it, so the result matches the reference; leaving the field out is not the same as auto. Then check the pixel dimensions of what comes back before you promise a client a 4K file.

Do not rely on target_pixels to reach a set size; treat it as not applied. Ask for the size through image_size or aspect_ratio on a model that supports it, and check the dimensions of what comes back. Also remember that the Flux 3 launch price carries a 50% API discount through October 8, which is a vendor offer and does not apply to Sume.

For the Flux side in more depth, see what Sume serves from Flux 3 and Flux 2.