Ideogram v4 Changed How It Bills — And Got Cheaper

Short version: Ideogram v4 stopped charging per image and started charging per megapixel, across three speed tiers. At the sizes most people actually generate, that makes it cheaper than v3. On Leaxor it is 1 credit an image, or 10 cents, at the QUALITY tier.

Disclosure: Leaxor is mine, and v4 went onto the picker today alongside v3.

What Ideogram is for

Ideogram has one thing it does better than almost anyone, and it is unglamorous: it spells.

Ideogram built its reputation on exactly this. Most image models treat text as decorative squiggles that resemble letters at a glance. Ask for a poster with a headline and you get confident nonsense. For thumbnails, packaging, ad creative and slides, that is the difference between an image you ship and an image you rebuild in an editor.

v4 does not change that positioning. It changes what it costs.

The billing change

v3 charged per image. v4 charges per megapixel, at three tiers, per the published endpoint rates:

Three speed tiers, each roughly triple the one below it per megapixel: TURBO, BALANCED and QUALITY.

This is a genuinely different shape, and most write-ups quote the cost of a 2048x2048 render as if it were the price of an image. It is the price of a large image.

Generate at 1344x768 instead, which is a standard 16:9 thumbnail, and you are paying for about 1 megapixel rather than 4.2. Same model, same tier, roughly a quarter of the cost.

Why that makes v4 cheaper than v3

A flat per-image price charges the same whether you render a thumbnail or a poster. Per-megapixel billing rewards asking for what you need.

Most real work is well under 2048x2048. Social images, thumbnails, blog headers and story frames all sit around 1 megapixel.

Which is why v4 is 1 credit, or 10 cents, on this platform while v3 is 2. A newer model that costs less is unusual enough to be worth stating plainly.

We request the QUALITY tier rather than the BALANCED default, and you are not charged extra for it. The model is sold here on text accuracy, and shipping the mid tier under that promise would be a quiet downgrade.

The other settings, and what they do

Rendering speed gets the attention because it drives the price. Three more parameters change the output in ways worth knowing.

Magic prompt expansion, set to Medium by default

The model rewrites your prompt before rendering, at None, Medium or Large. Medium is the default.

For a vague prompt this helps. For a prompt where you have specified the exact words that must appear in the image, an expansion step is a model rewriting your copy, and it is the first thing to disable when the text comes back subtly wrong.

Acceleration trades quality for speed

None, low, regular or high. Defaults to none, which is the correct choice for a model bought for text accuracy — the whole reason to use Ideogram is that the letters are right, and acceleration is the setting most likely to compromise that.

Output format

JPEG by default, PNG available. For images that are mostly flat colour and sharp type, which describes most Ideogram output, PNG keeps the edges clean where JPEG puts artefacts around every letter.

What Ideogram is genuinely for

Thumbnails with the headline baked in

The strongest case. Generating the image and the text together means the type sits in the scene, lit by the same light, rather than floating on top the way an editor overlay does. That reads as designed rather than assembled.

Posters, packaging and ad creative

Anything where a word is load-bearing. A beautiful product shot with a garbled label is worthless, and the cost of finding out is a reroll.

Mockups with real copy

Putting the actual headline into a layout mockup rather than lorem ipsum changes what people say in review, because they respond to the real thing instead of imagining it.

Where it loses

Images with no text in them. Ideogram is competent rather than exceptional at pure imagery, and its specialism costs you nothing but also buys you nothing there.

The honest rule: if the prompt contains a word in quotation marks, come here. If it does not, do not.

Where it sits

If your image is mostly text, v4 is the value pick. Nano Banana Pro earns its second credit when the prompt also demands precise composition, but paying double for text alone is not a good trade.

How to prompt for text that comes out right

Ideogram is better at text than almost anything else, and it still fails in predictable ways. Four habits fix most of them.

Put the exact words in quotation marks

Write the copy you want, in quotes, exactly as it should appear. Describing it loosely gives the model licence to paraphrase, and a paraphrased headline is a wasted render.

Keep it short

Reliability drops with length across every model that renders text. A three-word headline lands consistently. A paragraph will have a mistake in it somewhere. If you need a lot of copy, generate the image without it and set the type in an editor.

Say where it goes

Text placed by accident lands in the wrong third, over the subject, or half off the frame. Say it: across the lower third, top left with clear space around it. Composition instructions apply to type as much as to the subject.

Describe the lettering, not just the words

Heavy condensed sans, thin serif, hand-painted on a wooden sign. Without direction the model picks, and it tends toward a generic default that looks like every other AI poster.

The variants, and when to reach for one

v4 ships in more than one form. Fast and instant trade fidelity for speed, and both are aimed at iteration rather than final output. There is also a LoRA variant for teams training the model on a specific visual identity, which needs an asset pipeline most people do not have and is not something we run.

For almost everyone the base model at the QUALITY tier is the right answer, which is what runs here.

Should you use it?

Yes for anything with words baked in. Thumbnails with headlines, posters, ad creative, packaging.

Yes over v3, which is the same job for one more credit.

No if the image has no text and you want maximum aesthetic quality. That is a different model's job.

Ideogram v4 is live in Leaxor's image generator alongside nine others, credits at $0.10 and no subscription required. The Nano Banana comparison covers the other option for text-heavy work, and the August 2026 roundup has the rest of the month.

Reading this with an AI assistant? Ask it to summarize this in ChatGPT, Claude, or Perplexity.