Ideogram 4.5 edits: default medium, 5 references, vs GPT Image 2.5

On Sume, Ideogram 4.5 edits the first reference image and uses up to four more as guidance, five in all, with quality defaulting to medium; ChatGPT Image 2.5 takes up to 16 references and defaults to high. If a brief needs a handful of references and fixed per-quality prices, Ideogram is the simpler model; if it needs many inputs or a mask, GPT Image 2.5 is the one that lists them.

Both are edit-capable in the catalog, and in both cases an unlisted parameter returns 400 unsupported_parameter.

Side by side

Facts come from the Image API docs and the repository catalog; the Ideogram prices are the Fal list prices the docs quote, before Sume pricing.

Where the price shape differs

Ideogram's Fal list price is the same at every size: $0.03 at low, $0.06 at medium and $0.22 at high. ChatGPT Image 2.5 depends on size and quality: the docs give $0.09366 at xhigh and $0.21072 at max for a 1024x1024 output, before input tokens and Sume pricing.

That makes Ideogram easier to budget when you vary sizes, while GPT Image 2.5 prices follow output tokens.

• An Ideogram edit without aspect_ratio keeps the source image's shape.
• On GPT Image 2.5, use aspect_ratio: "auto" explicitly for the same effect.

Using the five references well

Ideogram 4.5 treats the first reference as the image to edit and the remaining four as guidance. Spend them deliberately: the base image first, then a logo, a style example, a color palette card and, if needed, one layout sketch. Five is enough for most poster jobs, and extra clutter makes the result harder to steer.

If you do hit the ceiling, merge references into one board image before sending. Two logos on one white card count as one reference, and a model that reads the card still sees both.

• Order matters: the edited image is always first.
• Use resolution: 2K only for finals.
• Keep the prompt specific about what to leave unchanged.

Which one for what

Choose Ideogram 4.5 for a poster where you send a layout and a logo and want a fixed price. Choose GPT Image 2.5 when you need a mask, 6 or more references, or exact custom pixels. Prices live in GET /v1/images/models; check the row before you commit.

Because the two models price differently, the best test is a small shared brief. Pick one poster, run it on Ideogram 4.5 at medium and on ChatGPT Image 2.5 at medium, and record the cost and the time to an approved result. Then run the same on a second brief with more references. If the Ideogram result is good enough on the first and the second needs more inputs than it accepts, you have a clear rule for which model gets which job.