- Imagine Weekly β The only Midjourney Newsletter you need
- Posts
- π Midjourney Office Hours Highlights: September 25th
π Midjourney Office Hours Highlights: September 25th
In the latest Midjourney Office Hours on September 25th, 2024, the team shared a bunch of exciting updates and projects they're working on. Check out the comprehensive summary of what was shared. If you prefer to just get the gist of it, jump down to the TL;DR; section.
Extended Recap
Here's a detailed recap of what was shared:
π§ Model Training: V7 Challenges
The V7 model training has been a complex endeavor, involving multiple individual components, each with its own set of bugs. The team has adopted a strategy of disassembling, retraining subcomponents, and then reassembling them for the main training. However, simultaneous changes have caused widespread issues. Reflecting on this, the team identified their biggest mistake: attempting to change everything at once instead of opting for incremental releases. Despite these challenges, there is optimism as individual components show improvements.
π₯ Video Models Development
The focus on training video models continues with weekly improvements. The team is considering releasing a V6 video model as a pre-alpha to collect data. They are debating between native direct flow resolution and an upscaling process, training multiple versions to find the optimal approach. The main question remains: when will the model be ready for release?
πΌοΈ 3D Model Progress
Daily collection of 3D data is underway, with larger 3D model tests anticipated to begin after a week of data collection. The 3D model could either be an extension of the video model or an independent entity. The expected release timing is before or shortly after the video model.
π Release Strategy and Lessons Learned
The V7 model is expected to take the longest due to its extensive changes. Early discussions on V8 have begun, focusing on separating releases into smaller, more manageable updates. The team is learning to manage multiple projects simultaneously, emphasizing the importance of a self-managing, cohesive team with strong organization. They are transitioning from bespoke events to formal processes with rhythmic release cycles.
π Psychometric Studies and Surveys
Positive data from psychometric surveys has been received, and the team expressed gratitude to participants. Plans are in place to publish reports on aesthetics and psychometrics. Upcoming visual personality tests aim to explore connections between aesthetics and personality, with additional personality question tests to gather more data and support scientific research.
π Storytelling Team Updates
The storytelling team is preparing its first demos and interactive releases in the coming weeks. Sessions will be conducted to showcase new tools and gather feedback. While acknowledging that these storytelling tools are experimental and not yet long-term products, the team intends to keep evolving them based on user feedback and experimentation.
π€ Personalization Efforts
The launch of personalization on the website aims to increase personalized model usage. Challenges remain in making the personalization flow easier and more intuitive. Debates are ongoing about messaging, flow improvements, and explanation clarity. The goal is to have personalized models as the default for most users, balancing automation and user control in the personalization setup.
ποΈ Editor Development and Moderation
A new image editor is in development and is expected to be a major upcoming feature. Plans are to lock editor features to higher-tier subscriptions for moderation purposes. The team is shifting towards pro yearly subscriptions to reduce abuse and streamline moderation. Continuous improvements to the UI and session management for the editor are also underway.
π Explore Page Enhancements
Recent algorithm changes aim to make the Explore page more diverse. While acknowledging that increased diversity may not always be better, the team is focused on improvement. Further enhancements are expected in the near future.
πΊοΈ Company Offsite in San Francisco
This week, the Midjourney team gathered in San Francisco for a company offsite, engaging in hackathons, nature activities, and project discussions. This in-person meeting aims to foster better collaboration among team members. While acknowledging a dip in productivity during the offsite, the team is optimistic about the creative outputs that such gatherings can inspire.
π§ Team Needs and Projects
The hardware team is looking to expand, with a need for more frontend developers. Various secret projects are in the works, though details remain undisclosed. These ongoing projects will be announced in the future.
TL;DR;
πΊοΈ Company Offsite: Team in San Francisco for hackathons, nature activities, and project discussions.
π§ Model Training: V7 training faces challenges; optimism remains with individual component improvements.
π₯ Video Models: Weekly improvements; considering V6 pre-alpha release for data collection.
πΌοΈ 3D Models: Daily data collection; larger tests to begin soon.
π Release Strategy: V7 takes longest; V8 to have smaller updates; focus on cohesive team and rhythmic cycles.
π Psychometrics: Positive survey data; upcoming reports and visual personality tests.
π Storytelling: Preparing demos; tools are experimental and evolving.
π€ Personalization: Website personalization launched; ongoing efforts to improve flow and user control.
π§ Team Needs: Expanding hardware team; secret projects in the works.
ποΈ Editor Development: New image editor in development; features for higher-tier subscriptions.
π Explore Page: Algorithm changes for diversity; further enhancements expected.
Stay tuned for more updates as Midjourney continues to innovate and enhance its platform!
If youβre not subscribed yet to my newsletter Imagine Weekly, Iβd be thrilled to welcome you on board!