Translate your comic's speech bubbles: one Ideogram 4.5 edit per page

Letter a 20-page comic in 3 languages with Ideogram 4.5 edits on Sume: one call per page, 60 calls, $2.25 at low quality, and why per-bubble calls cost 6x.

5 min readSume
All posts

To translate the speech bubbles of your own comic with AI, send each finished page as the first input_references image to POST /v1/images with Ideogram 4.5 and list every bubble in one prompt: number each bubble, give the old line and the new one. A 20-page chapter in three languages is 60 calls, which cost $2.25 at low quality ($0.0375 each on Sume) and $4.50 at medium.

This post is for creators lettering their own work, or work they hold the rights to translate. Running someone else's pages through an edit model does not make the translation yours. If your comic already has the lettering on a separate layer in a design file, use that file; this is the route for flattened pages.

Why one call per page and not per bubble?

Each Ideogram 4.5 call is one image and one charge, whatever the number of changes inside it. A page with six bubbles can be one call or six, and the price difference is exact.

20-page chapter, 6 bubbles a page, 3 languages. Sume list x 1.25; Image API list prices read 2026-10-05.
MethodCallsAt low ($0.0375)At medium ($0.075)
One call per page60$2.25$4.50
One call per bubble360$13.50$27.00

The per-bubble route is six times the spend and has a second cost: each call is another place for the art to drift. Ideogram's launch post says 4.5 "eliminates artifact buildup, making multi-turn editing possible" (Ideogram on X, read 2026-10-05). Even if that holds, fewer passes means fewer chances to find out it does not.

What does the prompt look like?

Number the bubbles by reading order, and say where each is, because a model cannot guess which "Okay." you mean on a page with three of them:

Edit this comic page. Replace the text in each speech bubble:
1 (top left bubble): "We have to go." -> "Tenemos que irnos."
2 (top right bubble): "Not without the map!" -> "¡No sin el mapa!"
3 (bottom caption box): "Meanwhile..." -> "Mientras tanto..."
Keep the lettering style, size and bubble shape. Do not change the artwork,
the panel borders or the sound effects.

What do I send and what comes back?

Use the page as the first reference. Leave out aspect_ratio, and the edit keeps the source geometry (per the Image API docs). Comic pages usually sit at 2:3 or 3:4, and both are in Ideogram 4.5's list if you ever want to set one. Sume takes public HTTPS image URLs, so upload the page first.

The partner page for the model (Morphic, read 2026-10-05) lists a 25 MB file limit and says the model translates text in place. A scanned page above that size needs a downscale first, and the Sume docs only promise that the reference is a public image URL.

What to check on every page

Do these four checks before you accept a page:

  • Count the bubbles. A missing bubble or an extra one is a fail, even if the text is right.
  • Check text fit. Spanish and German lines run longer than English, so shorten the translation before the bubble fills up.
  • Compare the line art. Faces, hands and backgrounds must match the source.
  • Keep sound effects. If a sound effect is part of the art, tell the prompt to leave it, and decide separately whether to retype it.

A trial before the whole chapter

Pick the worst page, the one with the most text, and run it at low. If one bubble is wrong, rerun at medium for $0.075. If that fails as well, break that page into two calls with half the bubbles each, which costs $0.15 at medium and still comes in under one high edit at $0.275. Use a key such as ch3-p07-es-v1 so a retry returns the same job and does not bill again.

What goes wrong first on a comic page?

Speech bubbles are a fixed size, and translated text rarely has the same length. A line that fit in English may overflow in German, so shorten the translation before you send it. Lettering style is the next thing to check: compare the font weight and slant against the source on every page, and look at the tails of bubbles, which should still point at the speaker.

Run a short test of two pages in the target language before you submit a whole chapter. It costs $0.075 at medium and tells you whether the prompt needs a rule about line breaks or capital letters.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume