Stability AI, a company that finances and distributes the Stable Diffusion image generation AI, has announced the update of the latter to V2. Among the novelties, we find the impossibility of generating adult content, to the chagrin of some users…
Earlier this year, the creators of Stable Diffusion regretted that their platform was being used to generate pornography, so they decided to address it.
A contested decision
Indeed, artificial intelligence has notably enabled certain users to create pornographic images resembling real people, as well as child pornography visuals. To prevent this from happening again, pornographic or nude images have simply been removed from the data used to train artificial intelligence.
However, unlike its rival Dall-E, Stable Diffusion is an open source model allowing people to improve the tool and developers to integrate it into their products for free. However, the open nature of AI is an argument to criticize the decision to ban adult content according to some users. On Reddita netizen said that this decision was “ censorship » and went « against the philosophy of the open source community “.
For the time being, it is unclear whether the filters added to Stable Diffusion to prevent the production of pornographic content will be effective; as another user recalls, the fact that the AI is open source means that the data it trains on can easily be changed in its third-party versions.
Stable Diffusion also tackles copyright infringement issues
Through its update, Stable Diffusion has also decided to tackle a second major problem posed by image-generating AIs: copyright infringement. Images that look too much like the work of certain artists are now “outdated” so that they can no longer be related to the style of the latter. This is for example the case of digital artist Greg Rutkowski, often used as a source of inspiration by users of this technology. In the past, Rutkowski has criticized the use of these artificial intelligences, like several online artistic communities which are fighting their democratization.
The V2 of Stable Diffusion brings of course improvements, in particular an increased quality of the images generated by the V1, as well as more advanced functionalities in general. The software can for example produce content corresponding to the depth of an existing image while being very different from it.
Sources: Engadget, The Verge
4