Updated
Ideogram 4.5: the model that writes legible text inside the image
Ideogram is kept here for one specific thing. It is the only connected image model whose letters come out as letters — signage, posters, logos, a cover with a headline. The others draw something that resembles text, and up close it always shows.
Below is the model itself, not a screenshot. Describe an image with a caption: the answer arrives on this page, and the first attempt needs no sign-up.
What you are trying here
This is not a demo mode: the request goes to the same model the main chat uses. The only difference is that the conversation ends after the first answer — you can continue in chat, and the image travels with you.
Text on the image — the reason it is switched on
The wording goes straight into the request, and best of all in quotes: “a poster with the words ‘Autumn Sale’”. The model places it in the composition and renders it so it can be read rather than guessed.
In practice that closes a whole class of jobs that used to need a designer or manual layout:
- Posters and playbills with a headline
- Signs, plaques, wayfinding
- Logos and wordmarks
- Covers, banners, headers
- Memes and cards with a caption
Do not bake long text into the picture
One or two short lines is the model’s strength. A paragraph in small type is not: the more characters, the higher the chance of a typo, and a letter cannot be fixed in a finished image — only regenerated.
The reliable trick for complex layout: get the background and the elements from Ideogram, and set the text itself as real text on top — then it stays editable and is guaranteed to be free of mistakes.
A transparent background in one request
The second thing the neighbours do not have: Ideogram returns an image with a genuinely transparent background if you simply ask for it in words — “sticker, transparent background, no background”. No separate mode to enable; the output is a PNG with an alpha channel.
This is the only source of true cut-outs in the system, and layered collage and poster assembly rests on it: each element is generated separately with transparency, then stacked. A composition built that way can be moved and corrected, unlike one baked in a single pass.
Cutting out someone else’s image costs pennies
The separate “remove background” operation takes any image — from another model or your own — and costs about three times less than the cheapest generation. If the object is already drawn and only needs separating from its background, regenerating it with transparency is noticeably more expensive than cutting it out.
Four operations and what they cost
You pay per image, and the price is set by the speed you pick. Changing an image you already have costs the same as drawing a new one, apart from the top speed.
| Operation | What it does | Relative price |
|---|---|---|
| Remove background | Cuts the subject out of a finished image onto transparency | The cheapest — roughly a third of a TURBO generation |
| Generate | An image from a description, at one of three speeds | TURBO is the base, DEFAULT twice that, QUALITY about three times |
| Remix | Changes only what you name in your image — a word, a colour, a detail — and keeps every other pixel; can also extend the frame to another shape | The same as generation; at QUALITY about twice a QUALITY generation |
| Edit | Combines up to five images by instruction: the first is changed, the rest are references | The same as remix |
Fix a detail instead of starting over
Got an image you like with one thing wrong — a letter, the sign colour? Ask remix to change just that. Everything else stays exactly as it was, and at TURBO the fix costs no more than a new draft.
A fresh generation from a sharpened description draws a different picture every time, so keep it for when you want a new idea, not a correction.
Extending the frame and rare shapes
Remix can also build out the edges: give it an image, ask for a different shape, and the model draws in what is missing rather than stretching. Plain generation already offers 4:5 and 5:4, along with elongated 2:1 and 1:2; from an existing image you can also go to 3:1 and 1:3 for banners and headers.
4:5 is the marketplace card shape, and Ideogram makes it directly. To get the same picture in several shapes, make it once and reframe it with remix rather than generating each shape separately.
How Ideogram differs from its neighbours
| Job | Which model |
|---|---|
| A legible caption, sign, logo or poster | Ideogram |
| A transparent background and cutting a subject out | Ideogram |
| Maximum resolution and a series of images | Seedream |
| Holding a face or a product across edits | Nano Banana |
What Ideogram will not do
- No 4K. An image is about 4 megapixels — 2048×2048 at 1:1. For a large format, make the image in Seedream and set the caption over it as real text.
- It will not guarantee flawless long text. Two short lines yes, a paragraph no.
- It will not hold a face across edits. That is Nano Banana.
- It will not return several images per call. There is no batch mode; options have to be requested one at a time.
Frequently asked questions
What is Ideogram for?
How do I make an AI write text on an image correctly?
Can I get an image with a transparent background?
How do I cut a subject out of an existing photo?
How do I change one detail without redrawing the whole image?
Can I make an image in 4:5 for a product card?
What is Ideogram’s maximum resolution?
Do I need to select Ideogram in the chat?
Examples and prices are on the Ideogram page and in image generation. Other models are in the book’s table of contents; images in general are in the image generation book.
Try it yourself
Everything in this guide runs inside Twelver
One chat for text, images, video, music and voice — no separate services or subscriptions.
Open Twelver chat