Stable AI releases version 2.0 of its Stable Diffusion AI image diffusion tool. The program has many new features. The update also removes the ability to create pornographic images and makes it more difficult to copy artists’ style, users report.
Stable Diffusion 2.0 gets, among other things, an update text to imagediffusion model Developer reports Stability AI. The program uses Laion’s OpenCLIP text encoder for this. This should “significantly” improve the quality of the generated images over previous v1 releases, the company claims. In version 2.0, the model can generate images with standard resolutions of 512 x 512 pixels and 768 x 768 pixels.
The program is getting a new upgrade tool, which can increase the resolution of images up to four times, say from 512 x 512 to 2048 x 2048 pixels. The update also adds a image depthFunction. This allows the program to create new versions of existing images while preserving depth information from the previous image. This feature can, according to Stability AI, create images that “look radically different from the original,” but preserve the coherence and depth of the original image. The program also has a feature that allows users to reuse parts of an image when creating new images.
Upscaler in Stable Diffusion 2.0 (left) and depth on image. Source: AI stability
However, the new version also caused a stir among users, The Verge writes, among others. For example, the new form gets a filter that prevents files from being generated Blow upPictures as pornographic images. Users Call change “oversight”although others point out that the option to create nsfw images May be added later with third party software.
Stability AI founder Emad Mostaki tells The Verge that he chose this because of security. “You can’t have kids and NSFW in an open model,” says the founder, referring to the ability to produce child porn with an open source tool. “So out with the kids or out with the NSFW.” It was also possible to create nude photos of specific people using the software without permission.
Source: AI stability
Users on social media are also complaining that it is difficult to use the artist’s style when creating images. Previously, users could use the artist’s name as a prompt when creating an image, but this seems to be less effective now. Artists like Greg Rutofsky, whose technique of creating images is often used with artificial intelligence, complained before About using AI generators for this.
Reddit user compares Stable Diffusion 1.5 with the latest version helps you create images in different art styles, such as those of Vincent van Gogh, Pixar, and Rutowski. This indicates that the latest version is significantly inferior in quality to this. Mostaque tells The Verge that copying such art styles isn’t specifically prohibited in version 2.0. He says the reduced ability to copy artists is due to changes in coding and the way the software retrieves data.
Stable Diffusion is an open source model that can be used to generate images based on text input. Users can run the software on a computer, unlike services like Dall-E, which are only available via the cloud. The software is supported by multiple GPUs with 8GB or more memory. the program Available for free on github. There is, too Online demos of the AI image generator are available.
