Nano Banana Pro 4:5 at 1K is 928x1152, not 1080x1350: the 1.16x gap

On Sume, Nano Banana Pro at the 1K tier with aspect_ratio 4:5 returns a native image of about 928 x 1152 pixels. An Instagram portrait is 1080 x 1350. The width gap is 1080 / 928 = 1.164x and the height gap is 1350 / 1152 = 1.172x, so the native frame is roughly 16 percent short on both edges. The Image API docs say exact 1080 x 1350 output comes from a documented post-step through the job's target_pixels, not from the model itself.

Nano Banana Pro costs $0.1875 per image at 1K and 2K on Sume, and $0.375 at 4K. Ten native 4:5 images at 1K cost $1.875.

The pixel arithmetic

The numbers below are the documented native size and the Instagram target. The ratio of 928 to 1152 is 0.8056, a bit wider than a strict 4:5 (0.8), which is why the docs describe the final size as a separate step.

Choosing between native and target pixels

If a social platform accepts a slightly smaller image, the native 928 x 1152 frame is fine and avoids any extra processing. If you need the published dimensions exactly, ask for the target size in the job so the post-step produces 1080 x 1350. Sume's docs describe the field as target_pixels on the job; confirm the exact request shape in the current API reference before you build on it.

The cost question is separate. The post-step is not itemized in the Image API pricing lines that the docs list, and the docs state that endpoint pricing lines are the amount Sume charges to your wallet, so read usage.cost on a test image to see the billed figure for your settings.

Other 4:5 options on Sume

Nano Banana 2.1 is $0.10 at 1K and $0.15 at 2K; confirm 4:5 in its supported_parameters before you rely on it. GPT Image 2.5 takes custom sizes in multiples of 16, so a 4:5 frame of 1088 x 1360 is valid (68 x 16 and 85 x 16) though it is not 1080 x 1350. Pick by what you need: Pro for its image quality, 2.1 for price, GPT for custom pixels.

A cost check for a 30-post month

Say you post a 4:5 carousel cover every day for 30 days and keep two options per day to choose from. That is 60 images. At Nano Banana Pro 1K the bill is 60 x $0.1875 = $11.25. At Nano Banana 2.1 1K it is 60 x $0.10 = $6.00. At GPT Image 2.5 low in a custom 4:5 size, the list price at 1K is $0.02475 per image, so 60 images are $1.485, though the low tier is meant for drafts rather than finals.

Those totals are output-only list figures with Sume's 1.25 margin. Reference images, if you pass any, add input tokens on token-priced models, and the catalog formula rounds billed amounts up to the cent. Use the first generated image of the month to compare usage.cost against your estimate, then lock the model and tier for the rest of the batch.

• Pro 1K, 60 images: $11.25
• Nano Banana 2.1 1K, 60 images: $6.00
• GPT Image 2.5 low 1K, 60 images: $1.485 (drafts)