Midjourney Office Hours (2024-12-04)
Category
- AI Image
Gen
- Midjourney
Date
2024/12/04
Summary 🍀🧸
V7 is scheduled to be released in January 2025. Video release is expected to be delayed. Storytelling is also delayed. Moodboard: Personalization to be applied as a set of images.
URL
https://x.com/blackowl777/status/1864496878114558025
URL 1
Empty
URL 2
Empty
Release
Coming Soon
Informer
Empty
Midjourney Office Hours (2024-12-04)
Original
Midjourney Office Hours 2024-12-04
(Source: GenIArt)
(Source: GenIArt)
V7 Model Status
•
January 2025 Release Confirmed: December release no longer possible
•
Model training progressing well
•
Team working on optimizing text dataset balance
◦
Challenge: More text can improve quality but may degrade visual output
•
V7.1 and subsequent versions in development
•
Will introduce new aesthetic and personalization systems
•
Planning slower but higher quality model variants
◦
Similar to Q2/Q4 approach
◦
Responding to user requests for higher quality options
Video Model Strategy
•
Key Decision Point: Affordable vs. Premium Model
◦
Community poll showed 50/50 split on preferences
◦
Three potential approaches under consideration:
▪
Affordable model with lower performance
▪
Premium model with highest possible quality
▪
Limited-scope model with balanced cost/performance
•
Business Constraints
◦
Current video models require significant investment ($1B+ range)
◦
ROI challenges due to Midjourney's self-funded business model
◦
Technology limitations make affordable, high-quality video currently unfeasible
•
Plan user ranking collection for video model training
Upcoming Releases
•
Personalization Profiles
◦
Multiple personalization support
◦
New "mood board" feature: personalization from a bunch of images
•
Storytelling Tools
◦
First experimental version
◦
Demo planned for next office hours
◦
Potential capacity issues expected at launch
Future Development
•
Personalization Improvements
•
Enhanced character references for V7
•
New “omni reference” system in development
◦
Will allow specific object/design integration
•
Big batch release (December vs. V7 integration)
•
Final video model approach selection
•
Release schedule and prioritization for 2025
Architectural Changes
•
Moving toward modular architecture
•
Separating personalization from model versions
•
Decoupling datasets from model versions
•
Independent scaling and architecture modifications
Translation
Midjourney Office Hours 2024-12-04
(Source: GenIArt)
(Source: GenIArt)
V7 Model Status
•
Confirmed for release in January 2025 : December release is not possible
•
Model training is progressing well
•
Team working to optimize text dataset balance
◦
Challenge: Using more text can improve quality, but may result in poor visual output
•
V7.1 and subsequent versions in development
•
New aesthetic and personalization systems to be introduced
•
Planning a slower but higher quality model transformation
◦
Similar to Q2/Q4 approach
◦
Providing high quality options according to user requests
Video Model Strategy
•
Key Decision Points: Budget vs. Premium
◦
In a community poll, the preference was split 50/50.
◦
Three potential approaches are being considered:
▪
Low-cost model with low performance
▪
Premium model with the best possible quality
▪
Limited scope model that balances cost and performance
•
Business constraints
◦
Current video models require significant investment (in the $1 billion+ range)
◦
Midjourney's self-funded business model leaves ROI challenges
◦
Due to technological limitations, low-cost, high-quality video models are currently unfeasible.
•
User Rank Collection Scheme for Video Model Training
Coming soon
•
Personalized profile
◦
Supports multi-personalization
◦
New “Mood Board” feature: Personalization from a collection of images
•
Storytelling tools
◦
First experimental version
◦
Plan a demo during the next office hours
◦
Potential capacity issues expected at launch
Future Development
•
Improved personalization
•
Enhanced character reference for V7
•
A new “omni-reference” system is under development
◦
Ability to integrate specific objects/designs
•
Massive Batch Release (December vs V7 Integration)
•
Selecting the final video model approach
•
2025 Release Schedule and Priority Setting
Structural change
•
Moving to a modular architecture
•
Separate model versions and personalization
•
Separate datasets from model versions
•
Independent extensions and architectural modifications