AI Speed Box

AI News and Information Link Collection (Mintbear's ignorant scraps, matured to become Visual AI News)
All
AI Image
AI Video
AI Sound
AI LLM
AI 3D
AI
AR, XR, VR
AI Toons
AI SNS

Midjourney Office Hours (2024-12-18)

Category
  1. AI Image
Gen
  1. Midjourney
Date
2024/12/18
Summary 🍀🧸
Recent updates Patchwork, Moodboards, Profile / v7 scheduled for January 2025 / Batch 8 capable of mass image generation in preparation.
URL
https://twitter.com/blackowl777/status/1869499353045324127
Release
Coming Soon

Midjourney Office Hours (2024-12-18)

Reference)
Midjourney is said to provide Batch for mass image generation, and Ideogram has just released BatchGeneration.

Original

Midjourney Office Hours 2024-12-18
(Source: JamesGriffing https://discord.com/channels/662267976984297473/1037743153471553618/1319048499441958982 )
Recent Feature Releases
•
Mood boards and multiple personalization profiles have been released
•
Encouragement to use custom models for improved results over baseline models
•
Introduction of an experimental research feature called "network" for world-building and storytelling exploration
Future Sharing and Exploration Tools
•
Plans to enhance exploration of S-REFs and mood boards for broader community sharing
•
A long-term goal: enable aesthetic exploration beyond all prior human history
•
Desire to coordinate community creativity to achieve an "aesthetic singularity"
Upcoming Batch Image Features
•
Consideration of "batch 8" features to manage and manipulate larger sets of images
•
Prioritization of mood board sharing before batch features
•
Ongoing debate on image resolution versus batch size optimization
Version 7 (V7) Model Development
•
V7 training is ongoing and may be ready by end of January
•
Focus of V7 is on character consistency and improved character references
•
Potential enhancements for style references, object references, and character-based storytelling
•
Exploration of optimal resolution, batch sizes, and upscaling strategies
•
After V7 release, plan to quickly iterate with follow-up models (eg, 7.1) and more frequent updates
•
Emphasis on decoupling architecture, data, and scaling releases to achieve a steady, frequent release cadence
Video Model Considerations
•
Current video models show mixed results
•
Trade-offs between quality, speed, and cost remain challenging
•
Plans to release some video functionality by January to gauge community interest
•
Evaluating whether to develop in-house video models, partner with third parties, or offer multiple options
•
Acknowledgment that high-quality video models may not be cost-effective yet
World-Building and Interactive Features
•
Long-term interest in world-building, storytelling, and immersive experiences
•
Possibility of future “walk-around” or real-time interactive features
•
Exploration of comic book-like storytelling and multiple character integration scenarios
Infrastructure and Server Capacity
•
Current surplus of server capacity due to early hardware arrivals
•
Introduction of a "holiday relaxathon" period to provide relaxed mode to all users
•
Intention to reduce or remove relax mode wait times, enabling near-unlimited image creation
•
Use of this period to gather data on server usage and community interest in high-volume generation
Emphasis on Fundamentals and Ongoing Improvement
•
Recognition that many users primarily use basic features rather than advanced tools
•
Commitment to improving core aspects: speed, resolution, quality, and prompt comprehension
•
Efforts to maintain a balance between new feature innovation and refining the basics
Aesthetics and Beauty in Model Output
•
Pursuit of models that not only produce realistic results but also more beautiful images
•
Continued encouragement for users to personalize models for better aesthetic outcomes
•
Plans to develop methods to improve overall visual appeal and engage more deeply with design aspects
Data, Scaling, and Release Cadence for 2025
•
New approach to model versioning to ensure more frequent and focused updates
•
Multiple planned models following V7 will incorporate different improvements (data, scaling, architecture)
•
Goal to continually learn from community feedback and usage patterns to guide future directions

Translation

Midjourney Office Hours: December 18, 2024

Recent feature updates

•
Moodboard and multiple personalized profile features launched
•
We recommend using a custom model
•
Introducing network features to support world-building and storytelling exploration: Patchwork

Future sharing and exploration tools

•
Plans to expand S-REFs and moodboard community sharing capabilities
•
Extending aesthetic exploration beyond human history
•
Aiming to achieve aesthetic singularity by mobilizing community creativity

Bulk image related features

•
Manage and review large volumes of images with Batch 8 functionality
•
Prioritize sharing mood boards
•
Discussing resolution and batch size optimization

V7 model development

•
V7 training in progress, possible release at the end of January
•
Improved character consistency and references
•
Enhanced style and object references
•
Review resolution, batch size, and upscaling optimization
•
Plans for rapid follow-up model releases and continuous updates

Video Model

•
Current video model quality is mixed
•
The trade-off problem between quality, speed, and cost
•
Some video features planned for January release
•
Consider in-house development, external collaboration, and multiple options
•
Lack of cost-effectiveness of high-quality video models

Worldbuilding and interactive features

•
Long-term interest in world building, storytelling, and immersive experiences
•
Exploring the possibilities of real-time interaction features
•
An attempt at comic book style storytelling and multi-character integration.

Infrastructure and server capacity

•
Secure server capacity
•
Provides relaxation mode to all users
•
Remove or reduce the relax mode wait time
•
Analyze server usage data and high-volume generation interest

Basic elements and continuous improvement

•
Improved speed, resolution, quality, and prompt comprehension
•
Maintaining a balance between refining core features and innovating new ones

Aesthetics of model output

•
Pursuing an image model that satisfies both realism and beauty
•
Improving aesthetic results through model personalization
•
Enhance visual appeal and explore design elements

Data, Expansion, and Release Cycles in 2025

•
A new approach to more frequent, focused updates
•
Integrating data, scaling, and architecture improvements
•
Adjust direction based on community feedback and usage patterns
👍
Made with Slashpage