Artificial Intelligence (AI)-based image and art generators are having a moment. While DALL-E 2 and Craiyon have garnered most of the attention so far, a new tool is now generating interest. It’s called Midjourney.
Go to Twitter and you will see hundreds of results of users who have experimented with Midjourney’s AI bot, which uses Machine Learning to generate images and “artworks” based on text prompts. Midjourney is gaining popularity on Reddit too, with gamers using it to explore artwork around games such as Dota 2 and Clash Of Clans.
These new AI tools let people create images through simple word descriptions. In July, AI research firm OpenAI had conducted a wide-scale beta test of DALL-E. “A video showcasing the tool on the company’s website showed DALL-E generating an image of a polar bear playing a guitar, a photo of a koala dunking a basketball, and the famed Mona Lisa painting but with the subject sporting a mohawk hair style,” AFP reported.
Midjourney describes itself as an independent research lab “exploring new mediums of thought and expanding the imaginative powers of the human species”. It has a small, self-funded team focused on design, human infrastructure and AI. Midjourney’s AI bot is currently available in beta, for use through the social platform Discord or Midjourney’s website.
Delhi-based podcast producer Aditya Worah, one of the few who has tried the Midjourney bot, says he came across it through other artists. “I was aware of the concept but I didn’t know what was creating it. You need a Discord account. Then you simply head to the Midjourney website. There you have an option to sign up. Every user gets 25 prompts. In a prompt, you could say, for example, a man sitting in a room, drinking coffee, in 60s aesthetics, looking at the sunset over New York,” says Worah.
The AI will start envisioning the image and present the results in close to a minute. Once it is ready, the bot also gives you the option of upscaling it to a higher resolution and saving it on your device.
Some of the prompts Worah used were “post-apocalyptic New Delhi in a cyberpunk style, 8K resolution” or “a woman wearing urban street fashion in Chandni Chowk in neoclassical Delhi”. Each prompt gives you four images or variants, says Worah. The results were fascinating. “The visuals it (Midjourney) used for Chandni Chowk were so cool. It almost looked like a dream,” he says.
Once you are done with the allotted 25 prompts, you will have to become a paid subscriber to generate more images.
As with every AI tool, there are concerns about violent content or images not suitable for viewing. Midjourney’s code of conduct clearly states the bot cannot be used to make “images that could inflame, upset, or cause drama. That includes gore and adult content.”
In July, OpenAI had said it had worked with researchers and developers to build in safeguards. “We reject image uploads containing realistic faces and attempts to create the likeness of public figures, including celebrities and prominent political figures,” Open AI had told AFP. Filters built into DALL-E also block violent, political, sexual or other content barred by its policies. In addition, the system is designed to avoid assumptions about race or gender.