On August 22 (local time), the UK startup company STABILITY AI has opened a high-performance image AI Stable Diffusion . Published for the general public. At the same time, the beta version of the web service DREAMSTUDIO using image generation AI has been released. Speaking of image generation AI, Midjourney, a service that allows you to create an image by entering a sentence to AI, has just become a hot topic on the Internet. However, some say that Stable Diffusion, which was released this time, is a higher-performance image generation AI.
Stable Diffusion can generate an image by entering a sentence to AI, like Midjourney. DreamStudio can be used by anyone if you register your account. Huggingface, a community site for AI engineers, has codes and documents, and there is also a demo page where you can try AI. It is also characterized by the fact that the distributed models are allowed to use regardless of whether they are profitable or non-profit by specifying the license. When I tried to enter only the word automaton using the demo page and output it, the following images were generated.
Since it is not a content that indicates a detailed situation, it is an image that only has an internal structure of a machine or a machine that imitates a human shape. With this alone, all the four images generated must be abstract. However, it takes a lot of time to create such an image by human hands. The emergence of a service that can automatically create such an elaborate image by just entering one word and waiting for a few minutes is a major technological innovation.
On the other hand, Stable Diffusion’s most specialty is a more realistic and photo-like image. By inserting appropriate texts, you can produce fictional animals that combine real animals and precise images such as the actual photographs that were actually taken. Mr. Yamakazu’s article, which explains the tips of how to use it quickly, lists a lot of images generated using Stable Diffusion. If you do not know that these are the images generated, it is inevitable that you will think that it is an actual photo.
It should be noted that Stable Diffusion has a very high accuracy of anime-style illustrations. Already on the Internet, illustration-like images created using Stable Diffusion have already been posted. It is a tremendous word that it is a quality illustration that seems to be a problem even if it is used as it is.
To generate more precise images, you still need a long sentence, and no one can call it easy. Apparently, the texts for generating the images they envision are tips. In addition, the name of spells seems to be established on the Internet that operates AI arbitrarily.
On the other hand, it seems that Midjourney, which has become a topic, is changing. The update that seems to have learned the STABLE DIFFUSION source has made it possible to draw two-dimensional characters here as well. It is an illustration-like image comparable to what was generated in Stable Diffusion. It is a unique feature of AI that it can evolve in a short period of time by incorporating open source.
There are also images that realistically describe the state of traffic accidents with fictional creatures using Midjourney. It is finished in a video like a movie scene, and is comparable to the video created in CG. It seems that the image generated AI has been demonstrated in this short period of time, regardless of illustration or live-action.
However, it is true that these image creation AIs are innovative and have several problems. The problem is particularly considered to be the handling of the generated images as copyrighted works. The use of copyrighted work in AI’s learning data is often interpreted as no problem. AI uses randomly learned data, but paintings, styles, compositions, etc. are not copyrighted. However, it cannot be denied that the generated image may violate the portrait right of a specific person.
Opinions are also exposed to whether copyright will occur to what AI has output. In some cases, images similar to existing illustrations are accidentally generated. However, proof that AI intentionally imitates the existing illustration is that the criterion is vague and very difficult.
In addition to Stable Diffusion and Midjourney, image generation AIs have already been released or are scheduled to be released. The appearance of image generation AI has had a variety of impacts in each industry, and a horror novel game using images created in Midjourney has already appeared (related articles). Although the issues to be cleared are left, it is expected that content using image generation AI will increase in the future. And with the appearance of the open source Stable Diffusion, the future that can easily generate high-quality images regardless of device may not be so far away.