Midjourney Unveils First Image Model in a Year, Soon After ChatGPT Launch

Midjourney has introduced its latest AI image model, V7, marking the first major update to its image-generation capabilities in nearly a year. The rollout began in alpha on April 4, shortly after OpenAI’s viral image generator debuted within ChatGPT.

Personalization and New Architecture

V7 is Midjourney’s first model to feature personalization enabled by default. To access it, users must first rate approximately 200 images to establish a personal visual profile. Once that step is complete, users can activate V7 on the Midjourney website or via the Midjourney Discord server.

According to CEO David Holz, V7 is built on a “totally different architecture” and brings significant improvements in prompt comprehension and visual coherence. Holz noted that the model produces more refined textures and shows noticeable advancements in rendering hands, bodies, and objects with greater accuracy.

Draft Mode and Model Options

V7 comes in two modes: Turbo and Relax. Turbo mode delivers faster image generation at higher computational cost, while Relax mode is more economical. The update also includes a new Draft Mode, which generates lower-quality images at ten times the speed and half the cost of standard images. These draft images can later be enhanced with a single click.

At launch, V7 will not include standard features such as image upscaling and retexturing, but the team expects to roll out these features in the coming weeks.

Early Testing and Community Feedback

Initial impressions of V7 suggest it handles text prompts effectively and maintains a strong level of detail. However, Holz cautioned users that the model may respond differently to prompting styles compared to earlier versions. Midjourney has asked users to experiment and provide feedback during the alpha testing phase, which will continue with updates over the next two months.

Founded in 2022 by David Holz, Midjourney remains independent and has not taken outside funding. The company reportedly projected $200 million in revenue for 2023 and recently revealed plans for a hardware team, though no further details have been disclosed. In addition to its work on image generation, Midjourney is developing models for video and 3D object creation.

Despite its growth, the company is dealing with legal issues related to the use of copyrighted images in training its AI models. Lawsuits claim the company scraped art from the internet without permission, raising questions about the data sources used in model development.