🚀 Midjourney Office Hours Highlights: January 22nd

The Midjourney team is really good at keeping us on our toes with the release date of V7. At least February is still on 👌 As the team gears up for the release of the V7 model, several exciting developments are on the horizon, promising to enhance user experience and expand creative possibilities.

For a quick overview, head straight to the TL;DR; section otherwise keep on reading.

Extended Recap

🖼️ V7 Model and Release Timing

The much-anticipated V7 model is undergoing final refinements, with a focus on addressing data quality issues discovered in the new dataset. This has necessitated a retraining using the V6 dataset, pushing the launch to February. The V7 model aims to offer enhanced multilingual support, improved coherence, and better character references. However, some aspects of image quality and aesthetics may still require further enhancements beyond this version.

🎥 Video Functionality

Midjourney is exploring two main directions for video functionality: maximum quality and maximum speed. The former promises higher-quality output but is slow and costly, while the latter offers faster generation at a lower quality. The team is weighing the potential costs to users, especially those on the $10/month tier, and is considering whether to develop an in-house video model or license external APIs. The likely approach is to offer both fast (in-house) and slow (licensed or external) models to gauge user interest and usage.

🖌️ 3D Exploration

Renewed interest in 3D capabilities has emerged following promising internal tests with updated datasets. The team is assessing the feasibility of launching an initial 3D feature soon, with decisions pending on the polish and effort required for these outputs. There is long-term potential to integrate 3D and open-world simulation experiences, which could revolutionize user interaction with the platform.

📊 Big Batch V6 (V6.2)

Before the V7 release, there may be an interim release of Big Batch V6 (V6.2), allowing larger batch sizes for improved personalization and control. Testing has shown significantly enhanced user control within each generation batch, although final interface and UX details are still being refined.

👤 Character References and Omni-Reference

The V7 model will address previous inconsistencies in character references, such as hands and weapons, and may introduce an "omni-reference" capability for incorporating specific items like logos and coffee mugs. This reflects strong community feedback emphasizing the need for character consistency and style control.

💸 Cost, Server Usage, and Community Impact

Budget constraints limit the feasibility of extremely expensive model training, such as $100M+ for certain video models. Large servers have been acquired in anticipation of new releases, allowing relaxed mode access for the $10 tier, though this may change with the launch of new features. The ongoing goal is to keep releases financially sustainable to maintain community support.

⏩ Multiple Model Tiers (Fast vs. Slow)

Observations indicate that most users prefer default or higher-quality (slower) modes, rarely opting for cheaper, lower-quality options. The team plans to explore whether a "fast" model for quick outputs or cheaper usage would gain traction. While professionals might pay more for higher quality, the broader community seeks affordability. Post-launch data on usage and subscription upgrades will inform future development priorities.

🔮 Future Focus and Platform Vision

Midjourney anticipates rapid improvements in image quality, speed, and workflow over the next two years. The team is considering adding advanced features such as 3D-based exploration, more flexible style referencing, and community-driven style libraries. Comprehensive solutions, including licensed external model APIs for video or other modalities, are also under consideration. Balancing multiple simultaneous projects, such as V7, 3D, video, and style tools, remains a challenge given the current team size.

🗣️ Community Feedback Priorities

Character creation, consistency, and advanced style controls remain top community requests. There is also a desire for additional collaboration tools, such as user-generated styles. Updates to the Explore page and potential new style discovery features are in progress. The team will likely focus on V7 first, then finalize these additional initiatives as resources become available.

TL;DR

🖼️ V7 Model: Launch delayed to February due to data quality issues; aims for better multilingual support and coherence.

🎥 Video Functionality: Exploring high-quality vs. fast, low-cost options; considering in-house vs. external APIs.

🖌️ 3D Exploration: Promising tests; potential feature launch soon, with long-term integration plans.

📊 Big Batch V6 (V6.2): Interim release may offer larger batch sizes for improved control.

👤 Character References: V7 to improve consistency; potential "omni-reference" feature.

💸 Costs & Community: Budget limits on expensive models; server capacity allows relaxed mode for $10 tier.

⏩ Model Tiers: Exploring fast vs. slow models; community seeks affordability.

🔮 Future Vision: Rapid improvements expected; considering advanced features and balancing multiple projects.

🗣️ Community Feedback: Focus on character consistency, style control, and collaboration tools.

Stay tuned for more updates as Midjourney continues to innovate and enhance its platform!

If you want to support me, feel free to buy me a coffee ☕️ 

If you’re not subscribed yet to my newsletter Imagine Weekly, I’d be thrilled to welcome you on board!