Users of AI image generator Stable Diffusion are unhappy with a major update to Version 2, which eliminated its ability to generate pornographic images as well as images in the style of specific artists.

Stable Diffusion’s developer, Stability AI, reworked key components of the AI ​​model in the update and improved certain features such as scaling (the ability to increase image resolution) and context-aware editing. However, the changes also make it harder for Stable Diffusion to work with certain types of images, which have caused controversy and criticism.

Users note that when asking Stable Diffusion Version 2 to generate images in the style of digital artist Greg Rutkowski, it no longer produces work that closely resembles his own.

Unlike competing models such as DALL-E by OpenAI, Stable Diffusion is open source. This allows the community to quickly improve the tool, and developers to integrate it into their products for free. But this also means that Stable Diffusion is less restrictive in its use and, as a result, draws considerable criticism.

In particular, many artists like Rutkowski are upset that Stable Diffusion and other image generation models have been trained on their work without any prior consent and can now reproduce their styles. Whether such copying by AI is legal remains an open question. Experts say training AI models on copyrighted data is likely legal, but certain use cases could be challenged in court.

Nudity and pornographic images were also removed from the Stable Diffusion training data. AI image generators are already being used to create NSFW output, including photorealistic and anime-style images. However, these models can also be used to generate images that resemble specific people (so-called “non-consensual pornography”), as well as images of child abuse.

While the changes in Stable Diffusion Version 2 annoyed some users, many others praised its potential for expanded functionality, such as the new ability for AI to create content that matches the depth of a real image.

In addition, Stable Diffusion 2.0 also includes the Upscaler Diffusion model, which increases the resolution of images by 4 times. Combined with text-to-image models, Stable Diffusion 2.0 can now generate images at 2048×2048 resolution and even higher.

The Stable Diffusion Version 2 update stopped the creation of porn and the generation of images in the style of specific artists

Stable Diffusion 2.0 upscales the generated low resolution image (128×128) to a higher resolution image (512×512)

Stable Diffusion Version 2 public demo is available here (although due to high demand on the user side, the model may be unavailable or slow).