Midjourney launches V7, its first new artificial intelligence model for images in almost a year.

Midjourney, one of the first web-based image generation services using artificial intelligence, has unveiled its new AI model after nearly a year. Named V7, this model began its alpha testing phase in the early hours of Thursday, just a week after OpenAI launched an image generator in ChatGPT that quickly gained popularity for its ability to create photos in the Ghibli style. While Midjourney's new model is not optimized for this style, it is capable of generating visually appealing works, at least according to this reporter's view.

The alpha testing phase of the V7 model has been noted as one of the most comprehensive to date, promising to be smarter, more beautiful, and more coherent than previous models. Users wishing to use V7 must first classify around 200 images to create a “personalization” profile in Midjourney, if they haven't done so already. This profile adjusts the model to individual visual preferences and is the first Midjourney model to have personalization activated by default.

Once the user completes this process, they can enable or disable V7 on the Midjourney website and, if they are a member of the platform's Discord server, also through the Discord chatbot. On the web application, it is straightforward to select the desired model from a dropdown menu next to the “version” label.

Midjourney's CEO, David Holz, described V7 as a "completely different architecture." In his announcement, Holz pointed out that the model is more effective with text prompts and that the quality of the images is visibly superior, highlighting the coherence and details in bodies, hands, and various objects. V7 is available in two versions: Turbo (more expensive to operate) and Relax, and it comes with a new tool called Draft Mode, which allows rendering images at ten times the speed and at half the cost compared to the standard mode. Although images in Draft Mode are of lower quality, they can be easily improved and re-rendered.

Holz also indicated that some standard features of Midjourney are still not available for V7, such as image enhancement and retexturing, but these are expected to be implemented in the near future. He emphasized that this is a new model with unique strengths and weaknesses, suggesting users explore different styles of prompts to discover its capabilities.

In preliminary tests, V7 showed reasonable compliance with prompts, although I did not have the opportunity to evaluate it in depth. Midjourney is a peculiar operation, founded in 2022 by Holz, who co-founded the PC peripherals company Leap Motion, without having raised external funding. By the end of 2023, it was reported that the company expected to generate approximately $200 million in revenue. Recently, the San Francisco-based firm announced the creation of a hardware team to tackle unspecified projects and continues to train previously announced models for video generation and 3D objects. However, the company faces multiple lawsuits accusing it of infringing the rights of millions of artists by training its AI tools using images pulled from the web without the consent of their creators.