Unveiling Ideogram 1.0

Ideogram introduces version 1.0, its latest text-to-image model. Developed entirely from the ground up and trained by the startup, Ideogram 1.0 showcases features in text rendering, photorealism, and the introduction of the innovative Magic Prompt. Let’s see what its all about and how does it compare with Midjourney?

Advancements in Text Rendering

In their own tests, Ideogram 1.0 says it addressed historical inaccuracies prevalent in AI-generated text within images. The model claims to establish a new industry standard by reducing error rates by nearly 2x in comparison to existing models. This leap forward simplifies the creation of personalized messages, memes, posters, T-shirt designs, birthday cards, logos, and more.

Image from the press release.

Showcasing Ideogram 1.0 Excellence

  • A captivating family portrait featuring a solid matte red sphere Christmas present, a blue cube, a vivid green Christmas tree, a curious dog, and an adorable cat.
From the press release
  • A cinematic, photorealistic portrayal of a rooster crafted entirely from crispy fried chicken, set in a dimly lit retro diner.
From the press release
  • Two men in a vintage study room, one standing with a disheveled appearance, the other seated in a formal suit. (we tested this as well versus Midjourney)
From the press release

In their own testing, Ideogram finds that people prefer how their image AI responds to the prompts, compared to Dalle3 and Midjourney:

From the press release

Introducing the Magic Prompt Feature

Ideogram 1.0 introduces the Magic Prompt feature, acting as a creative assistant that enhances, extends, and translates prompts. Examples include a humorous cat meme, a spooky Halloween scene, and a whimsical illustration of vegetables orbiting the sun.

What this does basically is turn your own text input, into a more detailed prompt. I assume this feature is to increase adoption for those users that are shy about being too descriptive with their prompts. If it works, it’s actually a great feature since one of the main pain points that people have with image AI is that they can’t express in words what they have in mind (shockingly!).

Here’s the example from the press release, and it surprised me particularly since we know how bad image AI’s are at text. We decided to test this as well.

Subscription Plans and Community Engagement

To access Ideogram 1.0 and be part of a global creative community, interested individuals can join ideogram.ai. The platform offers paid subscription plans with benefits such as priority generation, private generation, image upload, and access to the Ideogram Editor.

You get 25 image generations with the free plan, accesible here.

Securing Series A Financing

Ideogram has just secured $80 million in Series A financing. The funding round is led by Andreessen Horowitz, with notable participation from Index Ventures, Redpoint Ventures, Pear VC, and SV Angel. This is quite a good amount of money, given that Midjourney started with less and improved with just revenue from its user base. Unsurprisingly, Ideogram is now hiring to expand it’s team, so head on over to their website if you’re interested.

Testing – How does Ideogram look compared to Midjourney

Test 1 – Audrey Hepburn as Joker

I started looking at some of my recent Midjourney generated images, I guess at this point if you want to compete with the incumbent you have to go head to head with the current v6 version. I didn’t focus too much on some of the new features from Midjourney, like style reference because as I spoke before, this kills prompt engineering and it’s impossible to replicate across different tools.

Let’s look at what I did. First of all, this week Alibaba released a new image to video AI (Harry Potter anyone?) so I decided to use the exact prompt I used today for that featured image for the article.

The prompt is very vague, since I didn’t have any particular details I needed about the picture. While they do differ in style, I honestly see myself using at least a picture from either of them.

Test 2 – Photography test

If you’ve been following us, you already know we have an extensive photography guide for Midjourney – camera types, camera movements, angles, styles and on and on. I kept it simple, for another article this week I used a high shutter speed photography of a startup and wanted to see if Ideogram can match it. Even if the style is different, I would have had no issue using the first image it generated. It still delivers acording to the prompt and provides a reasonably good result.

Test 3 – Imagination

On our newsletter we have a daily Midjourney picture, this Tuesday we went with a surrealism photography. This tests the imagination of the image AI, since without any details it’s really up to it to decide what to include and how it makes the image look surreal. I think in this case Midjourney is really showing its mastery!

Test 4 – Clashing styles

I went with pictorialism for the photography style, which is a vintage style. I also added a rubber duck and I guess what makes a rubber duck a rubber duck is its shape and its color. So this mix should confuse the image generator a bit, since it has to blend black and white photography with a colored duck. In this case as well Midjourney wins.

Test 5 – Holograms

It really takes a bit of creativity to generate a different image for the same topic every day, and the most challenging for me is holographs. The picture still turned out good in Midjourney, although both don’t quite follow the prompt as much as I want— but in different ways. In this case Ideogram has certainly followed the prompt much better, but I have mixed feelings about the results.

Test 6 – Ideogram Use Case

All the previous tests was knowing Midjourney delivers good results against some prompts, and testing Ideogram’s ability to match it.

I was quite impressed with the image I saw in the press release, I decided to take the prompt and check it again in Ideogram and then see how Midjourney compares to it. The picture was presented in the press release to showcase the ability of the AI to follow the prompt and not ignore any part of it. In the case of the hologram I think it did just that.

In this case, I think both AIs did good, Midjourney possibly more polished. After doing this, it left me wondering if I tweak the text input a bit more, perhaps even in the previous tests Ideogram can match Midjourney. But that’s a test for another time.

Test 7 – Logos

After doing the tests, I looked at the interface and much like Midjourney, Ideogram shows you what other people have created. I noticed some logos that were surprisingly good. Not my cup of tea in terms of styling, but they were a very accurate representation of the prompt, down to the text level. I decided to use an existing prompt I saw (I won’t share the original, for privacy reasons as it was for a Telegram channel) and see what it thinks about our logo. In this case, Ideogram takes the crown by far as I don’t really understand what Midjourney was trying to do. I am flatterd though since I see a yin and yang representation there so there’s that.

Test 8 – Logos with Magic Prompt

In the interface, you can also select the Magic Prompt feature or not. I saw another logo in the public library that was even better than the first one, so I decided to reuse it for our tests.

I also activated the Magic prompt feature to see what it will do. What it does is turn your prompt from what you type in, to something a bit more detailed.

Here is the result. The prompt actually delivered something much better even on Midjourney.

Test 9 – Memes

After reading the press releasse, Ideogram was bragging about the ability of its AI to generate Memes. ‘That can’t be right’ I thought, or perhaps it’s only partially right. Nope, had to bite my own tongue.

Not only is Ideogram able to generate an actual good meme, it has the ability to come up with a joke, to write the text and to create speech bubbles if it wants to!

Instead of generating an image based on text, it actually shows evidence of being a multi model AI. This is similar to DallE, that sits on ChatGPT – it generates text but also images. This is very surprising.

Everyone knows this is the future of AI, a solution that can generate everything you want – text, image, video, sounds, software etc. It’s just a bit surprising that this pretty unique feature that seems like a competitive advantage is not advertised (or at least I couldn’t find any mention of it on their website).

What do you guys think about how these tools match up? Let me know in the comments.

While I’m already pretty much set in the Midjourney ecosystem, I think Ideogram delivers quite a good solution, considering they trained their model from scratch and this is just the 1.0 version. When we remember Midjourney is at version 6, it’s pretty exciting to think what Ideogram can achieve in a year or two.

Q&A Highlights

Q: What sets Ideogram 1.0 apart from existing models?

A: Ideogram 1.0 distinguishes itself through superior text rendering accuracy, photorealism, and prompt adherence, establishing a new standard for AI-generated images.

Q: How does Magic Prompt contribute to creativity?

A: Magic Prompt serves as a creative assistant, automatically enhancing and extending prompts to generate beautiful and imaginative images.

Q: What advantages do subscription plans offer users?

A: Subscription plans provide users with benefits such as priority generation, private generation, image upload, and access to Ideogram Editor, enhancing the overall creative experience.

Share.

I am equally passionate about technology, nature, ecosystems, and exceptional cuisine.

Leave A Reply

Exit mobile version