16 references on GPT Image 2.5, 5 on Ideogram 4.5: read limits in code
GPT Image 2.5 accepts up to 16 input_references and Ideogram 4.5 accepts 5 in total. A short Python script reads every model's reference limit from the catalog.

On Sume, GPT Image 2.5 accepts up to 16 input_references and Ideogram 4.5 accepts 5 in total: the first is the image it edits and up to 4 more act as references. The reference limit is part of each model's catalog descriptor, so read it from GET /v1/images/models instead of hard-coding a number.
Two limits documented in the Image API page
| Model id | References | What the first image does | Billed per image |
|---|---|---|---|
| openai/gpt-image-2.5 | up to 16 | edit target or reference; optional mask_url | 7 cents at high |
| openai/gpt-image-2.5-sunburst | up to 16 | same as Flare | 7 cents at high |
| ideogram/ideogram-v4.5 | 5 total | the first image is edited; up to 4 more are references | 8 cents at medium |
Read the limit for every model
Each model's supported_parameters includes an input_references range descriptor with min and max. This script prints the maximum for every row. A max of 0 means text-only.
import os, requests
r = requests.get(
"https://api.sume.com/v1/images/models",
headers={"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"},
timeout=30,
)
r.raise_for_status()
for m in r.json()["data"]:
ref = m["supported_parameters"].get("input_references")
print(m["id"], ref["max"] if ref else 0)Why the number matters
A reference-heavy edit can include a product shot, a logo, a hand pose and a background plate. With a 5-image cap you must merge or drop inputs, while a 16-image cap lets you pass them all. The extra input also counts toward GPT Image 2.5 pricing, since the docs state that admission includes estimated input tokens, so a 16-reference call can cost more than the 7-cent catalog figure.
Keep reference order stable between runs. The order is part of how the model is steered, and changing it is a common reason a rerun looks different.
Rules that apply to every reference
Reference URLs must be public HTTPS. Sume rejects localhost, private-network and non-HTTPS addresses before submission, so a signed URL on a private bucket works only if the host is reachable from the public internet. Each entry is an object of type image_url with the address under image_url.url. Ideogram 4.5 without references generates from text, and with references it edits the first image.
Plan a four-photo edit as an example. Four references on GPT Image 2.5 at high bill about 7 cents plus input tokens, and the same four on Ideogram 4.5 fit inside the 5-image cap, billing 8 cents at medium. A twelve-photo gift basket fits only on GPT Image 2.5. Write the limit check into your code once, so a thirteenth photo fails in your script instead of after you pay for a submission.
Sources
Related posts
More in Models
- 20 s Seedance: only 2.5 passes 15 s, $11.56 vs two 2.0 jobs at $7.56
Seedance 2.0 stops at 15 s on Sume; 2.5 takes 4 to 30 s. At 720p 9:16, 20 s on 2.5 is $11.56, two 10 s 2.0 jobs $7.56, one 20 s Wan 3.0 job $2.50.
- 3:2 photo ratio image API: GPT Image 2.5 does not list it on Sume
3:2 is the classic camera ratio, yet GPT Image 2.5 and GPT Image 2 do not list it on Sume. Fourteen other rows do. Table of rows and the crop workaround.
- 4:5, 9:16 and 16:9 together: 15 of 19 Sume image models list all three
Which Sume image models list 4:5, 9:16 and 16:9 in one catalog row, which four miss 4:5, and what each costs per image.
- 4K AI video on Sume: Omni Flash $3.75 or an H3 upscale $2.00 for 10 s
Gemini Omni Flash 1.1 lists 4K natively: $3.75 for 10 s on Sume. MiniMax H3 bills 4K only as an upscale from 768p, about $2.00. What to check before you pay.
Written by Sume