We are in the midst of an explosion of imagery.
by Kevin Convery
Although DALL-E and its successor DALL-E 2 get most of the headlines, the fact is we are in the midst of an explosion of imagery that has been generated via artificial intelligence from natural-language text input. Since DALL-E 2 arrived in September, excitement — and controversy — have followed the phenomenon around the internet. While the imagery generated by AI is certainly entertaining, and even useful, working artists have taken issue with the ability of online image generators to copy their style, deliberately and accurately.
Among the many alternatives to DALL-E, some are completely free, and some require a subscription. For this article, we’ll stick to the ones that are at least partially free.
The seven best alternatives to DALL-E 2
The more AI art generators you try, the better you’ll understand how the process works and what its capabilities, and limitations, are. Here’s a list of sites that we’ve tried that give results which measure up to DALL-E 2.
• Stable Diffusion
• Wombo Dream
Midjourney’s image generation goes on in a Discord room. When you go to www.midjourney.com, you’ll be invited to join the Discord room that serves as the beta test for Midjourney’s AI image engine. You enter “/imagine” in the chat box, and then your terms. Midjourney generates the typical four renditions as a message in the room, along with the choice to upscale any one or all of of them, or to create variations on each result. You can click on any image to download it.
Midjourney creates painterly textures very well, and the level of detail some artists are coaxing out of it is truly impressive. But the busy nature of the room means your results are continually scrolling out of sight as the art of other users comes up. It also gives you only 25 renders before you have to subscribe to make more.
Craiyon is very similar to DALL-E, on which it was based. In fact, Craiyon’s name was DALL-E Mini until OpenAI asked them to change it. Craiyon was trained on a smaller database than DALL-E was, so it is not as precise as DALL-E 2. Faces in particular look smeared and indistinct. But it is free and unlimited and the content range is unrestricted in regard to adult or political subject matter. Craiyon also provides a forum to share your results with the world.
Stable Diffusion is slightly different from the other image generators we’re discussing because you can download the code for it and run it on your own computer (you’ll need a dedicated GPU). It creates beautiful images of landscapes and architecture.
Stable Diffusion has real trouble creating photorealistic images of people or animals, which come out extremely distorted, but other art styles (such as watercolor or pen-and-ink) look like the product of a talented human artist. It has some of DALL-E 2’s restrictions (such as nudity) but is more permissive in other areas that DALL-E 2 restricts — go ahead and make pictures of political and other famous figures.
Dream by Wombo is a free AI art generator that is available on their website and as an app for Android and iOS. It is free to use, and the AI seems to specialize in dreamy, surreal images, but it can also create faces and human figures much better than some of the other choices.
Close-ups and portraits have much more defined details than longer perspectives. At Wombo Dream, you can augment your text input with a mandatory choice of art style. The output is locked to portrait orientation, unlike the square frames seen elsewhere. Nudity or sexual topics will be rejected by the app, but famous names and politicians are acceptable.
Simplified is a commercial website that offers various AI-driven products, such as computer-generated copy for blog posts. It has recently added AI-generated imagery. The images are meant for use in blog posts and similar online applications, so the images are small compared to DALL-E.
The interesting thing about Simplified’s UI is that you not only enter the subject matter as text, but also choose a style, a camera angle, and a filter (though you can leave these blank). Like many of the sites on this list, Simplified’s faces are much more recognizable in close-up perspective. Simplified has restrictions on both nudity and politics/famous people in place. It also restricts you to 10 generated images before you have to pay.
Starryai is similar to Wombo Dream and Simplified. You have a text input and a selection of styles to choose from. This engine is a bit slower; the product takes significantly longer to generate.
Starryai will let you upload a photo as a starting point for the image generation. It also seems to have the same problem as Stable Diffusion in that human faces and limbs come out distorted and unrecognizable in photorealistic renders — stick to watercolor or oil paint styles to get the best out of Starryai. For landscapes and moody backgrounds, however, it excels.
This entry is interesting in that in addition to the text input and a menu of art styles to mimic, ImgCreator.ai will let you choose your aspect ratio from square, portrait, or landscape. Like Starryai, it allows you to upload an image for the AI to get creative with, and the results are startling, as it filters your image in the same art style as you describe in text.
It is equally adept at creating landscapes or fantasy scenes as it is at human portraits and animals. ImgCreator.ai is only free for 10 images. As with many of the AI art sites, you can purchase more.
What can I do with the art I generate?
Anything you could do with something you physically created — print it, sell it, make a new product with it, etc.
Can an artist sue me if I make art in their style and sell it?
Currently, no. Some artists are concerned about this development, however, so what may be permissible in the future is anyone’s guess.
Is art made by a computer really art?
The answer to that depends on whether you see art as the final product, or the process by which it is inspired and created.
published November 22, 2022 at www.androidauthority.com