Change 2026 to 2027 on 30 old graphics: an Ideogram 4.5 edit batch

Refresh the year on 30 finished graphics with Ideogram 4.5 edits on Sume: 30 calls cost $1.13 at low, what the queue accepts per plan, and what to proofread.

5 min readSume
All posts

To change a year on finished graphics, send each graphic as the first input_references image to POST /v1/images with model: "ideogram/ideogram-v4.5" and a prompt such as: replace the text "2026" with "2027", keep the font and position, change nothing else. Thirty graphics are 30 edits, which cost $1.125 at low quality ($0.0375 each, Sume's list price of $0.03 times 1.25) and $2.25 at medium.

The job is dull and common: every December you have posters, headers, slides and badges that carry the old year. Re-exporting them from a design file is better when you have the file; this recipe is for the PNGs whose source is gone.

What goes in the prompt?

Quote the exact old string and the exact new one, and name where it is, since a graphic may carry the year twice: once in a headline and once in a footer. If a graphic has two occurrences, say so (both occurrences) or you may get one changed and one missed. Send no aspect_ratio, and the edit keeps the shape of the source, per the Image API docs.

Do not ask for other changes in the same call. A year swap is a precise edit, and each extra instruction adds a place where the result can drift.

How many edits will my plan accept at once?

Sume accepts valid paid jobs as queued while queue capacity remains, and refuses with 429 queue_full when both the processing slots and the queue are full. The Generation admission docs list the defaults, which also set how many of your 30 edits you can submit in one wave.

Default generation limits per plan from Generation admission, read 2026-10-05. The dashboard Concurrency tab is the source of truth for your workspace.
PlanProcessingQueueAccepted at onceWaves for 30 edits
Free1565
Pro420242
Startup840481
Scale201001201

A wave here means: submit up to the accepted count, wait for jobs to finish, then submit more. On 429 queue_full, back off and retry with the same Idempotency-Key. The errors docs say the same, and a retried submit with the same key and payload does not bill a second job.

What does the batch cost at each quality?

Ideogram 4.5 prices by quality, not by size, so a small badge and a 2K poster cost the same per edit.

Ideogram 4.5 at Sume list x 1.25. Image API list prices read 2026-10-05: $0.03, $0.06, $0.22 per image.
QualityPer edit30 graphicsRe-run 5 failures at the next tier
low$0.0375$1.1255 x $0.075 = $0.375
medium$0.075$2.255 x $0.275 = $1.375
high$0.275$8.25none

What to proofread

A partner page for the model (Morphic, read 2026-10-05) says Ideogram 4.5 edits stylized text in place and copies unchanged pixels from the source. Sume's docs do not promise either, so proof every file.

  • Small footers. A © 2026 line set at 10 pixels is the likeliest miss; rerun those at medium.
  • Other numbers. A 2026 edition number or a season label must not change. Name the one string to change.
  • Date ranges. 2026-27 is not 2027. Handle ranges by hand or with their own prompt.
  • The rest of the picture. Flip between source and result; anything that moved is a failed edit.

Send the 30 submissions with stable keys such as year-2027-<filename>. If your script dies at graphic 17, rerunning it replays the first 16 as the same jobs instead of billing them again, and the image-edit docs show every field you need.

How do I audit the batch afterward?

Keep a manifest with the source file, job id, prompt and a pass or fail mark per graphic. Read the year on all 30 at 100 percent, then spot-check three for anything that moved. Re-run only the failures with new idempotency keys; at low each re-run is $0.0375 on Sume.

A small test before the full run

Before you scale this up, run it on two or three real files first and write down what you saw. A small test at low quality costs cents on Sume ($0.0375 per Ideogram 4.5 edit), and it tells you whether your prompt, your source files and your review step are ready.

Keep the originals untouched, name every output after its source and its prompt, and store the job id with each result. If a result is wrong later, you can find the exact request, fix the prompt and re-run only that item with a new Idempotency-Key.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume