Ideogram is an impressive artificial intelligence image tool, but despite launching in August last year and having an uncanny text rendering ability, it hasn't garnered the attention of some of the other more high-profile GenAI services.
That is about to change as the company unveils version 1.0 of its generative AI model, capable of creating even more consistent, stylish and legible text on images.
The Toronto-based startup recently raised $80 million in its Series A funding round which it will use to accelerate its growth and build a wider range of media models.
I’ve been using Ideogram since version 0.1 and it was already impressive. The company says its new 1.0 model was "trained from scratch" with improved photorealism, complex prompt understanding and rendered text.
While you can already create rendered text with MidJourney v6, DALL-E 3, Stable Diffusion 3 and Google's Imagen 2 — the results are hit and miss.
What makes Ideogram stand out?
One of the big standout new features in Ideogram 1.0 is its Magic Prompt. This takes your short prompt and re-writes it to better reflect your intentions or vision.
For example you could ask for a funny meme about cats, and Magic Prompt would turn that into a much more descriptive overview — including describing each element of the cat, how the text should render and giving it a mood.
You can already do this to some degree with ChatGPT by having the OpenAI chatbot come up with a detailed prompt before sending it to DALL-E. Leonardo also offers a prompt refinement tool but with Ideogram it happens automatically.
However, the company says it has also worked on making it easier for the model to understand complex prompts and adhere to the request. So if you are a prompt engineer or particularly descriptive in your desires, Ideogram grasps the full request more accurately than before.
What about the text?
Version 1 includes an updated text rendering engine. This significantly improves the accuracy and legibility of text generated by the AI engine.
All of the images in this article were generated from prompts I inputted into Ideogram, specifically requesting text on the images.
Text on images only started to become legible in the big player tools like DALL-E, MidJourney and Imagen from Google late last year but Ideogram has had some degree of text capability since launch.
During some testing it seemed of Version 1, it seemed able to accurately render not just the main words but also a sub heading or promo line. It wasn’t always perfect but significantly more consistent than other models.
Ideogram says this increased accuracy makes it possible to "effortlessly create personalized messages, memes, posters, T-shirt designs, birthday cards, logos and more."
How do you access Ideogram?
Ideogram is a simple to use website with a prompt box, control options such as generated image size or style buttons and a selection of recently created generations.
With the launch of version 1.0 there is also a new subscription plan. While there will still be a free tier and daily free generations, for users requiring faster rendering and the ability to upload images as a source material — there are the paid options.
Prices for the premium plans range from $8 a month for 100 prompts per day to $20 for unlimited prompts.
There are no restrictions on usage regardless of whether you have the free or paid plan but on the free plan you can only download a compressed jpg file.
