If you've played with any of those text-based image generators in recent months, you know what I'm talking about. The open-source variants of OpenAI's DALL-E transformation model took the web by storm, and DALL-E mini became one of the most popular. However, the creators of the original DALL-E requested a name change to avoid confusion, and that gave rise to Craiyon, a new site that retains the same essence.
Creating art with artificial intelligence has never been this easy. You can visit Wombo Dream’s portal and generate the strangest landscapes, or go directly to Artflow.ai and get all kinds of portraits. What do these systems ask for in return? A bit of text, a description, or a random phrase. The algorithms handle the rest.
With that in mind, we come to DALL-E mini, which went viral in record time. DALL-E mini is an open-source edition of OpenAI’s DALL-E transformation model, and its team is already working on the second version. To avoid misunderstandings (and potential legal conflicts), OpenAI asked the creators of DALL-E mini to make a slight change to its identity, so the new site now goes by the name of Craiyon.
Craiyon: The Next Phase of DALL-E mini
Fortunately, there are no radical changes in how Craiyon works: enter a description, and within two minutes you should receive the first results. If the site reports excessive traffic, all you need to do is wait a few seconds and try again. The team has already announced additional servers, but requests multiply at certain times of the day.
The most important adjustment in Craiyon’s interface is the addition of a button to capture screenshots. The cropping of the images is exact, with a final size of 606 x 739 pixels. The portal’s FAQ explains that support for larger images won’t be available in the short term, but it could become reality later on.
Bonus Track: Latent Diffusion
Latent Diffusion works similarly to Craiyon, but it enables a series of advanced parameters: number of generation steps, width and height of each image, final number of images, and a “diversity scale”, which determines how different the images should be from each other. There’s plenty of room to experiment here, even if you decide to repeat a description. Give it a try!
Craiyon: Click here
Latent Diffusion: Click here
