Midjourney is in the final testing stages of V7, a significant update that rebuilds nearly every part of the AI image generator's system.
During this week's office hours session, developers explained that V7 features a new architecture, different data sets, and revised language processing approaches.
Internal testing could take a week to a month, followed by a two-week tuning phase and a potential alpha period with limited features.
“The V7 model is a complete overhaul of V6, including new architecture, different data sets, and revised language processing approaches,” Midjourney announced on its official server. “An alpha period often follows initial tuning, with limited features enabled.”
While V7's extensive rework continues, Midjourney's team is also exploring video generation capabilities through multiple approaches, including developing their own video model or partnering with external providers.
The main challenge, according to the team, lies in making video creation both user-friendly and cost-effective.
The team didn’t reveal a specific timeline for this video generator, so it may be in an earlier development phase. Right now, the developers are working on “balancing speed, cost, and output quality,” according to the announcement.
Around 85% of users prefer images generated using the platform's personalization system, which now includes mood boards and multiple profiles, Midjourney said. The company plans to expand these features by combining mood boards with style reference capabilities.
The platform also plans to introduce two distinct generation modes: a "real-time" option for rapid results similar to Meta’s “imagine” feature, Krea AI’s doodle-to-image feature, or Leonardo’s Realtime Canvas.
They also are working on a high-capacity mode for creating large batches of images similar to Leonardo’s Flow state.
Midjourney V6 was released in December 2023 and provided better results than its immediate predecessor, Midjourney V5.1, released in May 2023.
The results were noticeable, but the step from V5 to V6 was not as big as the upgrade from V4 to V5. Midjourney was more cautious when talking about V6, so its reasonable to think the team is very satisfied with the results to the point of calling this new version a “complete overhaul.”
During that time, the team also announced they were entering the AI hardware market but have yet to reveal further details.
Ever since the release of Midjourney V6, the team has expanded its capabilities, adding features like an image editor, a standalone interface, a style customization tool, outpainting and image expansion among others.
The team also started to pay attention to users’ feedback, sharing different polls to ask which features users want to implement.
Help us plan our roadmap over the next 12 months. Submit your ideas here: https://t.co/TuxV2p6UyI and we'll pool your thoughts together with our own and then put up a voting board to get a sense of what everyone would find most valuable. Thank you!
— Midjourney (@midjourney) November 1, 2024
According to Decrypt’s analysis, Midjourney currently excels in creative AI image generation, consistently producing highly aesthetic results that stand out from its competitors.
While it may occasionally sacrifice strict prompt adherence and text rendering accuracy, its versatility and overall capabilities make it a powerful tool.
Adding customizable styles has further enhanced its appeal, allowing users to create unique, personalized images that differ a lot from the platform's signature look.
Edited by Sebastian Sinclair