Image edit price sheet on Sume: Ideogram 4.5, GPT 2.5, cutout, upscale

Image editing on Sume is priced per completed job, and the four tools most people chain are: an Ideogram 4.5 edit at $0.0375, $0.075 or $0.275 per image by quality, a GPT Image 2.5 edit that depends on quality and size (from $0.009375 to $0.33375 at 1024x1024 with one input image), a background cutout at $0.0225 and an upscale at $0.20. Failed jobs are not billed. This post puts all of them in one sheet and prices three common pipelines.

Amounts for GPT Image 2.5 come from Sume's pricing test fixtures, which hold billed amounts for a 1024x1024 image with one input reference. The Ideogram prices are the provider list multiplied by 1.25, which the Image API docs describe as the billing rule.

Ideogram 4.5

ideogram/ideogram-v4.5 generates from text without input_references and edits with them: the first image is edited, up to four more are references. Quality is low, medium or high, the default is medium, and the price does not change with size. Ideogram calls 4.5 its most precise edit model on its own page. An edit with no aspect_ratio keeps the source geometry.

GPT Image 2.5

openai/gpt-image-2.5 accepts quality of auto, low, medium, high, xhigh or max, and defaults to high if you omit it. It takes up to 16 references, mask_url and background. Price depends on tokens, so it moves with size and with the number of input images. auto reserves the max price, so set the quality yourself.

Cutout and upscale

Sume RMBG 1.0 (POST /v1/rmbg-1.0/remove) takes a public HTTPS image_url, returns a PNG with alpha and costs $0.0225 per image regardless of size. Sume Image Upscale 1.0 (POST /v1/image-upscale-1.0/upscale) takes an image_url, an upscale_factor from 1 to 4 (default 2) and an output_format, at $0.20 per image, which is a reserved estimate of about 16 output megapixels.

Three pipelines, priced

Chaining is where costs add up. Here are three pipelines for 100 photos, with every step at the amounts above.

How to read the sheet

The upscale is the most expensive single step and often the least necessary: crop first and enlarge only what you will publish. The cutout is cheap and exact, so it beats a model for transparent PNGs. The edit models differ by control: GPT Image 2.5 gives you a mask, Ideogram gives you a fixed price per image and keeps the source shape. Always read live prices from GET /v1/images/models/{id}/endpoints, since the catalogue is the source of truth and this sheet is dated 2026-10-07.

Finally, note that these are list amounts for the date shown, so treat the tables as a planning aid and the endpoints route as the source of truth when you write a budget into code.

Budgeting tips

Set a ceiling per job in your own code and check the usage.cost field of each response against it. It holds the billed amount in USD, so you can total a batch without reading an invoice. Explore at the lowest quality that still shows the idea, and spend on high only for finals.

Poll long jobs through the jobs API rather than resubmitting; a timeout on a sync call returns a 202 job envelope, and sending the same payload again with a new key creates a second job. Use the same idempotency key and payload if you must retry.