Ideogram 4.5: what it costs, how Precise Edit works, and who sees your images

The Morse teamPublished November 12, 20268 min read

Ideogram 4.5 is Ideogram's image generation and editing model, built for changing one part of a picture many times over without the rest of it drifting. As of November 12, 2026, it runs in the Ideogram app, through its MCP server, and on its API, where an image costs $0.008 to $0.22 depending on the quality mode and the endpoint. Ideogram calls it "the most precise edit model." Unlike Ideogram 4.0, it has no public weights.

We read Ideogram's Ideogram 4.5 and 4.0 model pages, its homepage, plans and pricing page, licensing page, Ideogram 4.0 technical post, Ideogram 4.5 product photo guide, API overview, API setup guide, the API references for both Ideogram 4.5 endpoints, its terms of service, API agreement, and privacy policy, all on November 12, 2026. None of those pages gives a launch date. The earliest dated 4.5 page we found on Ideogram's site is its product photo guide of October 14, 2026, and Recraft's partner post of October 1, 2026 says Ideogram "released Ideogram 4.5 this week" (Recraft). Ideogram earns credit in its API agreement, which says in plain words that it won't train on what developers send.

How much does Ideogram 4.5 cost?

From under a cent to 22 cents an image on the API, set by the quality mode you pick and whether you send images in. The Ideogram 4.5 model page lists four modes on two endpoints:

The 4.0 column comes from Ideogram's 4.0 model page. At low and medium the two models cost the same. At high, any 4.5 job that starts from an image costs more than twice what a prompt does, so ten rounds of high-quality edits on one product shot come to $2.20, and the same ten rounds at low come to $0.30. The Generate reference sets quality to high for a prompt alone and medium when you attach images, and very_low works only with source images.

Does resolution change the price?

Ideogram's own pages disagree. The model page says each image "has a fixed price. It does not change with prompt length, the number of reference images, or the output resolution." The Generate reference says "Pricing is tiered by the resolved output pixels: up to 1024x1024 bills as 1K, above that as 2K. An 'auto' size bills as 2K." The Precise Edit reference matches the model page for edits, saying quality "is the only field that affects the price." Both endpoints take a dry_run parameter that prices a request without running it, and nothing is "generated, stored, or billed," so ask for a quote before a batch job on Generate.

API credit is prepaid. Ideogram's API setup guide takes one-time top-ups from $1 to $300, caps the balance at $300, and pauses your keys when it reaches $0. If you also pay for an Ideogram subscription, API calls spend the subscription's monthly credits first when they cover the whole request.

What does Ideogram 4.5 Precise Edit do?

It changes what you describe and copies everything else from your original. The Precise Edit reference says the output "always matches this image's width and height, and pixels the edit did not meaningfully change are copied exactly from it." You can add a mask, black for the area to change and white for the area to keep, and up to four reference images to steer it, or three when a mask is attached.

An edit of the whole image is capped at 4,194,304 pixels, a 2048 × 2048 square, so a larger image comes back scaled down: an 8192 × 8192 file returns at 2048 × 2048. Set context_window to auto with a mask, or name a region by its corners, and only that region is regenerated while the rest of a file up to 8192 × 8192 comes back untouched at full size. The model page shows this on a 4,016 × 6,016 photo, 24.2 megapixels, for colorways and print work.

Ideogram's model page compares 14 rounds of the same edits on Ideogram 4.5, GPT Image 2.5 Sunburst, Nano Banana Pro, and Nano Banana 2, and says the others "become unusable within a few edits." That comparison is Ideogram's own; we didn't run one. The Generate endpoint also rewrites your prompt by default. Under magic_prompt, the default auto and on both "rewrite and expand the prompt before generation," and off keeps your wording.

Is Ideogram 4.5 open source?

No. Ideogram's homepage heads its model section "Ideogram 4.0/4.5: The open models for visual intelligence," but only 4.0 has public weights. Ideogram's licensing page says "We are releasing the weights for Ideogram 4.0," offers them free under a non-commercial agreement or at $300 a month for a 10,000-image self-hosted commercial license, and says nothing about 4.5. Ideogram's 4.0 technical post of June 3, 2026 describes a 9.3-billion-parameter model whose nf4 checkpoint fits on a single 24 GB GPU. If you need to run it yourself, that is the Ideogram model you can download. Ideogram 4.5 is hosted only, in Ideogram's app and API and on partner platforms such as Recraft's.

Is Ideogram 4.5 free?

Only in small amounts. Ideogram's pricing page says "Eligible free accounts receive weekly credits when signing in with Google, Apple, or Microsoft; the amount may vary." Its product photo guide says Precise Edit "requires a paid plan or purchased credits," and the pricing FAQ says AI editing features need priority credits and aren't available in the slow queue. As of November 12, 2026, Plus costs $20 a month, or $15 billed yearly, with 2,400 priority credits, and Pro costs $60 a month, or $42 billed yearly, with 7,500. Pay as you go is 1 cent a credit.

Are your Ideogram 4.5 images public?

In the app, yes by default. The pricing FAQ says "Images are published by default to encourage sharing within the community" and that "Plus and Pro subscribers can generate unpublished images and unpublish previously published images," with deletion also limited to Plus and Pro. The same page's comparison table marks "Private generations" as included for pay-as-you-go accounts, which aren't Plus or Pro. If you're counting on that, check the setting before your first upload.

Ideogram's privacy policy, last revised January 15, 2025, says a generated image is "publicly available on the Services" alongside "your account handle and account picture," and its terms of service, last revised August 14, 2024, let other users remix anything that isn't Private Content. For how that default compares with Midjourney's and Leonardo's, see our Midjourney alternatives comparison.

Does Ideogram train on your prompts and images?

On the API, Ideogram says it won't, except for flagged content. In the app, its privacy policy allows it. Ideogram 4.5's examples start from uploads: product shots, room photos, and old family pictures to restore.

The API clause sits in Ideogram's Developer API Agreement, last revised August 14, 2024, which says that despite anything in the terms of service and privacy policy, Ideogram "shall not use any User Input or User Output to train the Ideogram AI Model," except input or output "flagged as violating the Company's Usage Policies," which it may use "consistent with the Company's mission of detecting abusive and unsafe uses." The app's privacy policy, which counts your prompts and uploaded images as User Input, makes no such promise.

How to use Ideogram 4.5 with your images kept on your device

Morse's Studio lists Ideogram 4.5 among its image models, so you can use it from a Morse account, with no Ideogram account or API key of your own.

• Results stored in your browser, encrypted. Studio images, whether a product shot or a launch visual, are stored in your browser, encrypted, and locked when you sign out.
• No training on what you make. Morse doesn't train AI on your content, sell your data, or show ads.
• Your plan. Ideogram 4.5 is among Studio's image models; check the model menu for your plan. Free gives 3 images a day with Z-Image Turbo and needs no card, and Pro, at $19 a month or $17 billed yearly, adds unlimited use of Pro image models such as Nano Banana 2 Lite, Qwen Image 3 Pro, and four unfiltered image models.

Studio images are stored in your browser, so clearing it deletes them. Keep a copy of anything you'd hate to lose.

Ideogram's homepage pitches 4.5 on prompt fidelity and "readable type in every layout," so the prompt and the words you want on the image do most of the work. Write them in a Morse chat, in the mode that fits:

• Private, to write a structured prompt with an open model. DeepSeek, Kimi, or GLM drafts the JSON prompt or the scene description, and Morse has zero data retention agreements with its inference, search, and other providers.
• Anonymous, to have GPT, Claude, or Gemini translate the headline or label text you want in the image. The provider gets the text without knowing it's you, and may keep it for a limited time to check for abuse, under its own policy.
• E2EE, for a product name or launch line nobody should read yet. Your browser checks the enclave's proof before anything leaves your device, the enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed. E2EE chats can't make images, so paste the finished prompt into Studio yourself.

Every chat in every mode is stored only in your browser, encrypted. E2EE comes with Pro.

Start free and open Studio's model menu to find Ideogram 4.5, or compare it with Seedream 5.0 Pro and FLUX.2 pro first.