Poster headline in the image: Ideogram 4.5 or Nano Banana 2.1 on Sume?

For a poster with a headline inside the image, both Ideogram 4.5 and Nano Banana 2.1 are on Sume, both vendors point to text rendering as a strength, and the catalog gives a cost ladder to test with: Ideogram 4.5 at $0.0375, $0.075 or $0.275 by low, medium and high quality, Nano Banana 2.1 at $0.075 to $0.20 by tier. Neither claim is a benchmark, so the right move is a small test on your own headline.

Vendor text claims come from pages I read on 2026-10-10: Google's image generation docs say Nano Banana 2.1 produces legible, stylized text for infographics, menus, diagrams and marketing assets, and the fal Ideogram V4.5 page shows accurate text rendering in its example prompt. Sume's side comes from the Image API docs and catalog code.

Side-by-side limits

Ideogram 4.5 takes up to 5 references (one image edited, four as references), a quality of low, medium or high (default medium), a resolution of 1K or 2K, and 15 aspect ratios including 3:1 and 1:3. Its output format list is empty, so you omit output_format. Nano Banana 2.1 takes up to 10 references, four resolution tiers, 15 ratios including 8:1 and 1:8, and png, jpeg or webp.

A 30-poster test in dollars

Thirty posters at Ideogram 4.5 medium cost 30 x $0.075 = $2.25. At low it is 30 x $0.0375 = $1.125, and at high 30 x $0.275 = $8.25. On Nano Banana 2.1 at 1K the same 30 are $3.00, at 2K $4.50.

A sensible test runs three headlines on each row, each at the cheapest tier that you would ship. Count how many images have every letter right. That count is the only evidence that matters for your own typefaces, languages and line lengths.

Traps to avoid

Do not send output_format: "png" to Ideogram 4.5, since the row lists no formats and the docs say the request is rejected. Do not send quality to Nano Banana 2.1, which does not list it. Use resolution on both rows, but remember the values differ: Ideogram 4.5 has no 512 or 4K tier.

• Quote the headline in the prompt and keep it short.
• Draw tiny legal text in code, not in the model.
• Fix the typo by editing: Ideogram 4.5 treats the first reference as the image to edit.
• Pin both rows; do not use sume/auto for a text test.

How to score the test

Write the pass rule before you generate. A strict rule is: every character in the headline is correct, the spacing is even, and nothing extra appears. Score each image pass or fail, and keep the failures, because they show whether a row breaks on a particular letter pair or on long lines.

Run each prompt several times, since a single lucky image proves little. Compute cost per passing image rather than cost per image: if a $0.0375 row passes half the time, each pass costs $0.075, the same as a $0.075 row that always passes. That figure, not the sticker price, tells you which row is cheaper for your headline.

Both rows are billed per output image, so asking for four variants in one call multiplies the price by four. Check the charge as the endpoint price times n before a big run, and keep the prompt wording identical across rows so the comparison stays fair. Save every output with its row id so you can reopen the comparison later.