Ideogram API Guide: Text-in-Image Generation, Pricing & Code (2026)

If you have ever tried to get a general-purpose image model to put clean, readable text into a poster or logo, you already know the pain. Most models mangle letters, warp spacing, or give up after a word or two. Ideogram built its reputation on solving exactly that, and it is still the model people reach for when words have to come out right. Here is what the API does, what it costs, and how to call it.

What Ideogram Is

Ideogram is an AI image model best known for rendering accurate, readable text inside generated images: logos, posters, signage, and anything where typography carries the design. That is a different core strength from general-purpose models like Midjourney or Flux, which are built around scene composition, style, and photorealism rather than getting the word SALE to actually look like the word SALE.

That focus makes it a common pick for design work driven by code: marketing teams generating ad variations, agencies mocking up packaging concepts, or apps that let people make their own branded graphics.

Worth knowing before you start: Ideogram has grown well past text-in-image. The API now covers vector output, transparent backgrounds, background and object removal, upscaling, reframing, and splitting a design into editable layers. Text rendering is still the reason most people arrive, but it is no longer the whole product.

Key Features

Text that comes out readable. This is the headline feature. Where other models treat lettering as one more visual pattern to approximate, Ideogram renders coherent characters, words, and short phrases, which makes the output usable as design rather than concept art you have to redo the text on afterward.

Style settings instead of long prompts. Ideogram supports style types (Realistic, Design, or Auto) and a long list of named presets, from Watercolor to Pop Art to Bauhaus, so you can steer the look with a setting rather than a paragraph of prompt wording on every request.

Editing an image you already have. Beyond generating from scratch, Ideogram can remix an existing image with a new prompt, repaint just the part you mark with a mask, change the shape of an image while keeping its content, and swap the background out. These all bill at the same rate as a normal generation, so editing is not a cheaper path than regenerating. Our guides to AI image editing APIs and using reference images cover the patterns in more depth.

Keeping a character the same across images. Ideogram has a character reference mode that takes one image of a person and holds that likeness across new scenes. It costs noticeably more than plain generation, roughly $0.10 to $0.20 per image on 3.0 depending on the speed setting, but it is one of the stronger options for this. Our guide to keeping AI-generated characters consistent compares it against the alternatives.

Real vector output. Ideogram Vector 1 generates SVG files from a prompt or from reference images, around $0.13 to $0.20 per file depending on quality, and will convert an existing image to SVG for about $0.10. This matters for logo work, where a regular generated image is a fixed grid of pixels that goes blurry when scaled up, and a vector file does not.

Access and Pricing

The API is billed and managed separately from Ideogram's consumer subscription. Billing is per generated image rather than per request, so a call that returns four images is charged four times the per-image rate for whichever model and speed setting you chose.

Ideogram publishes a full rate card rather than leaving you to guess, so these are list prices rather than estimates. On 4.0, Turbo is $0.03 per image, Default is $0.06, and Quality is $0.10. The newer 4.5 generation lines up with that: Low at $0.03, Medium at $0.06, and High at $0.10. Older models are cheaper, with 1.0 Turbo at $0.02 and 2a Turbo at $0.025. Prices do change, and Ideogram commits to 14 days notice by email before any material change, so check the live page before building a cost model on these.

One quirk worth catching: on 4.5, sending source images with your prompt changes the price in both directions. The Very Low setting with source images drops to $0.008 per image, the cheapest route in the whole lineup, while the High setting with source images jumps to $0.22, more than double the $0.10 you pay without them. If you are feeding in references at volume, that gap is the number to watch.

Three practical notes:

• There is no free API tier. Ideogram's free weekly credits apply to the app, not the API, which needs a payment method and a positive balance before any request will go through. Pay-as-you-go credits are a cent each.
• The default limit is 10 requests running at once. That is the ceiling that usually bites first on batch work, not a per-minute cap.
• If you would rather not run a separate Ideogram subscription and key, a platform that carries several models is a reasonable alternative, particularly if two or three other models already handle other jobs in your product.

Calling the API

Ideogram's own API expects an Api-Key header and a JSON body holding your prompt and settings, sent to the endpoint for whichever version you are using. The request shape differs between versions, so check the reference for the one you are targeting.

If you would rather not maintain a separate integration just for Ideogram, Apiframe carries several Ideogram versions, including a dedicated character model, behind the same endpoint it uses for Midjourney, Flux, GPT Image, and dozens of others. Our getting started guide covers getting a key. Here is a generation request:

curl -X POST https://api.apiframe.ai/v2/images/generate \
  -H "X-API-Key: afk_your_api_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "A poster with the text OPEN LATE in bold retro lettering",
    "model": "ideogram-v3",
    "ideogramParams": {
      "aspect_ratio": "16:9",
      "rendering_speed": "default",
      "style_type": "Design",
      "style_preset": "Bauhaus"
    }
  }'

Or in Python:

import requests

response = requests.post(
    "https://api.apiframe.ai/v2/images/generate",
    headers={
        "X-API-Key": "afk_your_api_key_here",
        "Content-Type": "application/json",
    },
    json={
        "prompt": "A poster with the text OPEN LATE in bold retro lettering",
        "model": "ideogram-v3",
        "ideogramParams": {
            "aspect_ratio": "16:9",
            "rendering_speed": "default",
            "style_type": "Design",
            "style_preset": "Bauhaus",
        },
    },
)
print(response.json())

The rendering_speed value is what sets the price, so switch it to the fastest setting while you are iterating on prompt wording and raise it only for the version you intend to keep. Model names and available settings vary by version, so confirm the exact strings on the model's documentation page before you ship.

The request returns a job ID straight away with a QUEUED status, since generation runs in the background. Check GET /v2/jobs/:id until it turns to COMPLETED, and the result carries your image URLs:

{
  "images": [
    "https://cdn2.apiframe.ai/images/a1b2c3d4-e5f6-7890-abcd-ef1234567890-1.png"
  ]
}

Those links are hosted for 90 days, so download anything you plan to keep rather than treating the URL as permanent storage. Rather than checking in a loop, you can have the job call your server when it finishes, which is the better pattern once you are past testing.

Errors are straightforward: 400 means something in your request body failed validation, 401 means the key is missing or wrong, and 402 means you are out of credits. They all come back in the same { "error": ..., "details": {} } shape, so you do not need separate error handling per model.

Timeouts need different handling from those three. A timeout means the response never came back, not that the job never started, so resending can quietly run a second generation and bill you for both. Send an idempotency key so the repeat is recognized as the same request. At a few cents an image this is minor, but it adds up on batch work.

Best Use Cases

Logos and branded graphics. Because the lettering renders correctly, Ideogram is one of the few models you can point at a prompt like generate a logo with the company name in it and get something close to usable first time. One caveat for real brand work: a standard generation is a fixed grid of pixels, so it will not scale cleanly to a billboard or a favicon. Either use Ideogram's vector output, or look at the Recraft API guide, since Recraft is built around vector and brand assets from the start.

Posters, ads, and social graphics with text in them. Anywhere a headline, a call to action, or a price needs to sit in the image itself rather than being added in a design tool afterward.

Fast iteration for marketing teams. Generation is quick and driven by the prompt, so a team can run through a dozen text and layout variations before settling on one. At $0.03 an image on the fastest setting, a dozen variations costs about 36 cents.

How It Compares for Text

Flux and Midjourney are both strong general-purpose models, but text has historically been a weak spot for both. Short words sometimes come out clean, longer phrases tend to distort or drift into nonsense. The gap has narrowed, though, and Ideogram no longer has this to itself. GPT Image 2 handles moderate amounts of text well, particularly for interface mockups and simple labels, and Nano Banana Pro now has the best text rendering in its family. Ideogram still tends to win on longer phrases and precise typography, and it is cheaper per image than either, but this is worth testing with your own text rather than taking on reputation. For pricing and quality detail on the alternatives, see the Flux API guide and the complete Midjourney prompt guide.

You can also get an API key and start with 100 free credits if you want Ideogram in the same integration as Midjourney, Flux, and dozens of other models.