Midjourney, one of the pioneering AI image-generating services online, has introduced its first new AI image model in almost a year. This model, named V7, began its rollout in alpha around midnight Eastern on a Thursday, following a week after OpenAI launched a new image generator in ChatGPT that swiftly gained attention for producing Ghibli-style images. Although Midjourney’s model is not officially optimized for Ghibli-style images, it still manages to generate visually appealing works.
To utilize V7, users are required to evaluate approximately 200 images to create a Midjourney “personalization” profile, unless they have done so before. This profile customizes the model to align with the user’s unique visual preferences. V7 marks the first instance where Midjourney has enabled personalization by default. Upon completing this process, users can activate V7 on Midjourney’s website and, for those in the Midjourney Discord server, through its Discord chatbot. The model can be selected using a drop-down menu next to the “version” label within the web application.
David Holz, CEO of Midjourney, described V7 as having a “totally different architecture” in a post on X. He elaborated further on Discord, stating that V7 is significantly more adept at interpreting text prompts and produces impressive image prompts, higher image quality, and improved coherence in various details like textures, bodies, hands, and objects.
V7 is available in two versions, Turbo and Relax, with the former being more expensive to operate. It also supports a new feature called Draft Mode, which generates images at ten times the speed and half the cost of the standard mode. While draft images are of lower quality than those in the standard mode, they can be enhanced and re-rendered with a simple click.
Some features, such as image upscaling and retexturing, are not yet available with V7, according to Holz. He mentioned these functions are anticipated to be introduced within the next two months. Holz emphasized that the model has unique strengths and potential weaknesses, encouraging users to experiment with different styles of prompting to discover its capabilities.
Midjourney, founded in 2022 by Holz, who also co-founded Leap Motion, has not relied on external funding. As of late 2023, the company was reportedly projected to generate around $200 million in revenue. Recently, Midjourney announced the formation of a hardware team to pursue undisclosed projects and continues developing previously announced models for video and 3D object generation. The company is also facing several lawsuits alleging infringement on artists’ rights by using web-scraped images without the creators’ consent.