Ideogram 4.5 has two endpoints; on Sume it is one model id

Ideogram 4.5 reaches you through two vendor endpoints, but on Sume there is one model id: ideogram/ideogram-v4.5. What you send decides what happens. No references means text-to-image. References mean an edit of the first image. The Image API docs describe the model that way, with up to 4 more references (5 in total).

How the two map

The two endpoint names come from the partner page Morphic's Ideogram 4.5 page, not the vendor, read 2026-10-05: precise edit, where the output matches the input size, and generate plus edit. The launch post on X (read 2026-10-05) says the model is live in Ideogram, the API and with launch partners. The Sume column is from the Sume docs.

A request for each case

Generate, no references:

POST /v1/images
{"model": "ideogram/ideogram-v4.5", "quality": "low",
 "prompt": "A tall poster with the words OPEN STUDIO in bold type"}

# Edit: image first, no aspect_ratio
POST /v1/images
{"model": "ideogram/ideogram-v4.5", "quality": "low",
 "prompt": "Change OPEN STUDIO to OPEN HOUSE. Keep everything else identical.",
 "input_references": [{"type": "image_url",
   "image_url": {"url": "https://example.com/poster.png"}}]}

Why this is worth knowing

Knowing the vendor has two endpoints helps when you read Ideogram's own docs or a partner's page, since their parameter names and limits belong to their endpoints and not to the Sume request. For example, a 25 MB file limit and a 1:6 to 6:1 aspect range appear on the partner page; Sume lists its own resolution values of 1K and 2K. When the two disagree, the Sume catalog is what your request will meet.

The practical consequence

You do not choose an endpoint on Sume, so you cannot pick the wrong one. You can still get a surprise: if you leave out the reference, you get a new picture, not an edit. If you send aspect_ratio on an edit, you ask for a different shape, and the docs say that an edit without it keeps the source shape.

Limits that stay

quality is low, medium (the default) or high, and resolution is 1K or 2K. The docs list mask_url for ChatGPT Image 2.5 edits, not for this model. The vendor-side mask support the partner page mentions is not a Sume parameter today. A parameter the model does not list returns 400 unsupported_parameter.

The vendor post also says open weights are coming soon. Nothing about the Sume request shape depends on that. Read GET /v1/images/models/ideogram/ideogram-v4.5/endpoints for the current parameters and the pricing lines before you build.

When you write documentation for your own team, say which endpoint a prompt was written for. A prompt that worked as a generate request may behave differently as an edit, since an edit starts from an existing image and the instruction must say what to keep. Label saved prompts as generate or edit, and keep the source image link beside each edit prompt. That small habit saves a lot of confusion later. Keep the scope of this advice in view. It rests on the Sume docs and the vendor pages named in the sources, read on 2026-10-05, and on nothing measured by Sume. Where a behavior depends on your own images, such as how a model redraws a certain typeface, run a small pilot at the low quality tier and judge the result yourself before you plan a batch. Write down the prompt, the model id and the quality tier you used, so the run can be repeated. When the catalog or the docs change, re-read them; the live catalog is the contract, and a post is only a snapshot of it.