An updated Stable Diffusion XL 1.0 neural network has been released - this is a cool analogue of Midjourney

Stability AI has announced the launch of Stable Diffusion XL 1.0, its most advanced text-to-image model ever. According to the creators, it offers brighter and more accurate colors and works with contrast, shadows and lighting better than previous versions.

Generation in version 1.0 has become much faster: it is reported that a picture in a resolution of 1 megapixel is now created in seconds. Text generation has also been improved: while most text-to-image models cannot create realistic inscriptions and logos, Stable Diffusion XL 1.0 is capable of "advanced generation text".

In addition, the new version offers painting (finishing the missing parts of the image), painting (expanding an existing picture) and working with converting a picture into a picture. The latter works much like a generative fill in Photoshop: the user selects an image and describes what changes they want to see in the output.

Another important innovation is that the neural network began to understand complex instructions in the form of short queries, while the previous version required large and specific descriptions.

The new version of Stable Diffusion XL is already available on two platforms owned by Stability AI: clipdrop And dream studio. You can try out the neural network on both for free, but picture-picture generation is available only through DreamStudio. There are also disadvantages: the finished illustration will have a watermark, and the queue for generation can be thousands images (to be fair, during testing, a queue of 1200 images took less than two minutes). Without registering, you can only make a few requests, then you have to create an account or wait 24 hours to update attempts.

Stability AI also announced a new partnership with Amazon. With this, Stable Diffusion XL 1.0 will be released on Amazon Bedrock, a cloud-based platform for generative AI models.