Nano Banana 2.1 or GPT Image 2.5 for product photos on Sume

For product photos on Sume, pick GPT Image 2.5 when you need a transparent cutout, a mask or more than 10 references, and Nano Banana 2.1 when you want a fixed price per tier and a wider ratio list. Both take reference images, and both are called through the same POST /v1/images.

How do the two rows differ?

The first four rows come from the Sume catalog code and docs on main. The last row is the vendors' own pricing pages.

Which one for a white-background catalog shot?

Either. Ask for a plain white backdrop and give the product photo as an input_references item. Nano Banana 2.1 is simpler to budget: $0.10, $0.15 or $0.20 billed by tier. GPT Image 2.5 bills by quality and size, so run the endpoint line for your settings first.

Which one for a transparent cutout?

GPT Image 2.5. Set background: "transparent" and a format that carries alpha. Nano Banana 2.1 rejects background with 400 unsupported_parameter.

{
  "model": "openai/gpt-image-2.5",
  "prompt": "Isolated wireless earbuds case, studio lighting",
  "background": "transparent",
  "output_format": "png",
  "aspect_ratio": "1:1"
}

What does GPT Image 2.5 cost?

The image models docs give a 1024 by 1024 example: xhigh output is $0.09366 and max is $0.21072 at list, before input tokens and Sume pricing. Lower qualities cost less. The Sume billed price adds the 1.25 factor, so read the endpoint line rather than adding it yourself.

A fair test costs little. Use the same product photo, the same plain prompt and a 1:1 frame, run each model once, and compare label text, edge quality and how well the product shape is kept. Keep the reference photo sharp and evenly lit, because both models copy its flaws. Decide on that sample, not on the matrix alone.