In the ever-evolving realm of artificial intelligence, image generation has been one of the biggest leaps forward this year. OpenAI has recently unveiled DALL-E 3, and there is a considerable leap in image quality compared to previous versions. While DALL-E 3 is a big step in quality for OpenAI, the text-to-image field is becoming increasingly populated.
MidJourney and Leonardo are at the forefront, carving significant niches for themselves. Their capabilities and refinement in image generation offer captivating visuals in a variety of styles. But the likes of Stability AI and Adobe are challenging in this space. Then, the shadow looming over all of this, are tech giants Google and Meta, who have both announced various forays into the text-to-image space to be released in the coming months.
With all of this in mind, it would be good to explore some of these tools, to help understand the kind of image quality that is possible right now. Let’s start by looking at the image quality difference between DALL-E 2 and DALL-E 3:
Prompt: “Greek ruins under water”
DALL-E 2 images often have a haze over the top of them, making them look more dreamlike. This has historically made it relatively easy to spot an AI generated image. However, DALL-E 3 looks to add more detail and realism to the images.
Now let’s explore that same prompt, but compare images across various other AI text-to-image platforms:
MidJourney, version 5.2
Leonardo, model “PhotoReal”
DALL-E 3
Stability, version SDXL, style “photographic”
Adobe Firefly
Now let’s explore what images look like of people, which is still one of the hardest aspects for AI to create seamlessly. Using the same 5 AI platforms, with this prompt: “A photorealistic portrait of a 25-year-old American girl with long, flowing blonde hair and striking blue eyes, wearing a white t-shirt, golden-hour sunlight, 85mm lens“. The Adobe results were a little disturbing, so have been omitted from this.
Leonardo, model “PhotoReal”
MidJourney, version 5.2
DALL-E 3
Stability, version SDXL, style “photographic”
Finally, let’s look at something a bit different. Animation is a relatively new option within AI. While photorealism still appears to be the measure of success, animation style is rapidly growing in usage. Let’s try the prompt “a wolf standing by a lake at sunset, animated“
MidJourney, version Niji V5
Leonardo, model “3D Animation Style”
DALL-E 3
Stability, version SDXL, style “anime”
Adobe Firefly