Leading AI image generator Stable Diffusion has announced the open-source release of Stable Diffusion Version 2 and has annoyed some users by removing the ability to copy the style of specific artists.
The new artificially intelligent (AI) model allows for larger images to be generated and has an added “upscaler diffusion” model that allows users to generate images of 2048×2048 or even higher. Typically, text-to-image generators only generate images of 512×512.
Stable Diffusion’s parent company Stability AI says that the new version “greatly improves” the quality of the generated images compared to the first version and has provided some impressive examples.
However, the new update has generated some consternation among its user base who have noted that the new version has removed the ability to generate pictures in the style of specific artists.
In its press release, Stability AI says it has removed the ability to generate NSFW images thanks to a new adult content filter placed on the software, but did not directly comment on the removal of specific artists in text prompts.
Arguably the most popular artist used to generate AI images is Greg Rutkowski whose name had become shorthand for producing fantastical high-quality images.
As reported by The Verge, users have taken to Reddit to complain that “they have nerfed the model.”
“What did you do to Greg,” commented one user. “It’s kinda an unpleasant surprise,” added another on Stable Diffusion’s official Discord server.
However, Rutkowski is presumably pleased with the latest development after telling Forbes that he found the amount of work being churned out using his name “terrifying.”
“Right now, when you type in my name, you see more work from the AI than work that I have done myself, which is terrifying for me. How long till the AI floods my results and is indistinguishable from my works?” He said.
Stability AI’s founder Emad Mostaque tells The Verge that it has not removed artists’ images from the training data but instead changed the way the program encodes and retrieves data.
“There has been no specific filtering of artists here,” Mostaque says.
Not all the feedback has been bad with some users noting that the new version is “better at interpreting prompts and making coherent photographic images.”
There is also a new feature that gives users the ability to produce content that matches the depth of an existing image.
Called “Depth-to-Image”, it can deliver transformations that look radically different from the original but still preserve the coherence and depth of the original image.
Leading Image Generator
Stable Diffusion is an important AI image generator because it is open source allowing developers to integrate it into their products for free.
The company plans to make money by training private models for customers and acting as a general infrastructure layer.
For example, genealogy website MyHeritage this month announced an AI Time Machine tool that allows users to create images of a person in different periods throughout history using Stable Diffusion’s technology.
A demo of the new Stable Diffusion version can be tried here.