Midjourney AI Office Hours Recap - July 17, 2024
Let's get up to date on what's going on at Midjourney.
Every Wednesday, the Midjourney AI team hosts a live event to discuss the product's future, with the CEO joining to answer questions.
You can tune in on Discord; here's a link to next week's office hours.
Today, I'll give you a brief overview of what was discussed in yesterday's office hours and what we can expect in the future. This is something new for the newsletter, and I'd love to hear your thoughts. Should I keep doing this?
Let’s get to it!
Key Topics Discussed:
Website Rooms syncing with Discord channels
New Improved personalization coming
V6.5 is getting closer to being finished
Testing runs for V7 started
A test version for depth control nets
Some 3D capabilities should be released this year
Working on style explorer
New “Editor” user interface planned
Working on implementing the ability to draw
Building a hardware team
Now, let's dive deeper into the most exciting highlights!
The Website
Currently, the website is in its alpha state and is only available to people who have generated over 100 images. To check if you are eligible, you can use the /info command on Discord.
I would highly encourage anyone who has access to the website to check it out. The user interface is far superior to Discord's. It’s much easier to use images, style, and character references, and to iterate on your generations.
With Rooms soon syncing with Discord channels, the website will also cover the social aspect of the platform.
I was used to using Discord, but after spending some time on the website, I can clearly say there’s no going back for me.
However, there are still some features missing, like the ability to set preferred options. This feature would allow users to create custom parameters that can be easily added to prompts with a single word instead of typing everything out.
For example, you could create a preferred option --portrait that would automatically expand to something like portrait photography, studio lighting, shot on Canon EOS R10 --ar 2:3.
Personalization
I’m currently working on a more in-depth article on the topic, but in short, personalization is a parameter you can add to your prompts that adjusts the outputs based on your personal aesthetic preferences.
A new and improved version of personalization is planned to be launched soon. This update will better unlock your personal preferences and allow you to save states of personalization as it improves.
Currently, personalization changes as Midjourney gathers more data on your preferences.
In the future, by combining this data with survey information on demographics, Midjourney could roll out models personalized for specific demographics.
New versions
The team is also working on new versions of the model. V6.5 is expected to be released in the next couple of months. They have already started working on the upscaler for V6.5. This version could come in two separate variants: one focused on image quality and another focused on perfecting details like hands. While the style of V6.5 will be similar to V6, the improvements will be noticeable.
The team has also started test runs for V7. This will be a more significant update, coming later, and will be much more different than V6.
Additionally, work is being done on 3D and video models. Some 3D capabilities might be released this year, while video will probably take longer. The CEO stated that they want the video feature to be consistently good, unlike the current offerings on the market.
Control Nets
One of the most requested features from pro users is control nets. The upcoming depth-based control nets will enable users to create variations of the same generation while maintaining the original composition. This feature will provide more precision and flexibility, allowing users to experiment with different aspects of their images without altering the overall structure.
Style Explorers
Last week, we discussed style references. Currently, to find the most fitting styles, you either need to use external references or generate images with random --sref until you find what you like.
There is an update in the works that could improve this process, making it easier to get the style just right. However, we don't have many details about this update yet.
New “Editor” interface
The team is also working on a new editing interface for the website. This interface will combine inpainting and outpainting capabilities into one window, allowing you to drag around the borders of images and use selection tools, like a brush, to change specific parts of the image.
In the future, we might be able to edit external images using this interface. However, for this to happen, a better moderation system needs to be put in place. Moderation is easier to manage for generated content than for inputs from outside sources.
The ability to draw
There are plans to add a new feature that allows you to draw and turn your drawings into AI-generated images. This feature aims to appeal to artists and address criticisms that AI art only requires typing in a few words. By integrating drawing capabilities, the platform hopes to offer a more hands-on and creative approach for users.
Closing words
That’s it for today!
I’d love to hear your feedback on these office hour overviews. Should I continue providing them weekly? Feel free to ask any questions—this newsletter is for you! If there are any specific topics you’d like me to cover, make sure to leave a comment.
