3D Optimism | Midjourney Office Hours Recap April 3rd 2024 | Midjourney News
TLDRIn the Mid Journey office hours recap, updates on the platform's development were shared. The team is focusing on new social features, personalization, and improving text and image accuracy. A caption party is planned to enhance the model's understanding of image-language connections. There's optimism about a high-quality 3D model in development. The team is also considering user feedback and potential new user roles for content moderation.
Takeaways
- 📝 Medium is a recommended platform for creatives, offering customizable prompts to potentially save time at work.
- 🏖️ Progress has been slower due to vacations, but the team is working on new social features for the website.
- 🤖 Testing of social features will begin with a limited number of spaces to stress test the system before expanding access.
- 🎨 Personalization features are in development but are progressing slower than desired due to the challenges of working across multiple time zones.
- 🔄 Style 'random' is expected to return, although details are not yet clear.
- 👥 An algorithm for improving hands, bodies, and text accuracy is in the works, aiming to reduce the frequency of poor image outputs.
- 🖼️ Efforts are being made to enhance image quality, particularly to address small pixel artifacts.
- 🚀 A potential speed update may increase efficiency by 25-50%, but it's contingent on completing other updates first.
- 🎉 A 'caption party' is upcoming, aiming to teach the version 7 model the connection between images and language, with possible future rewards.
- 🏆 The feedback leaderboard on the Mid Journey website will receive regular updates and community ratings to prioritize feature development.
- 🎭 The possibility of a new class of trusted users for rating and captioning is being considered, potentially linked to rewards.
Q & A
What is the main purpose of the Medium website mentioned in the script?
-The main purpose of the Medium website mentioned is that it sells customizable prompts, which can be useful for creatives to save time at work.
What is the current status of the progress on the website with new social features?
-The progress has been slower than usual due to people being on vacation. The main focus is on the website, including the new social features, which will be tested with guides and mods.
How many social spaces are initially planned for the new features?
-There won't be too many social spaces initially. The plan is to start with a low number of spaces with lots of people to stress test the system.
What is the team working on regarding personalization?
-The team is working on personalization features, but it's moving slower than desired due to having people working across multiple time zones.
What is the status of the style, random feature?
-Style, random will show up again, but the details are not clear. It seems like it will come from dial tuning, but users won't have access to the tuning part.
How is the team addressing issues with hands and bodies as well as text accuracy?
-The team is working on an algorithm to improve the accuracy of hands, bodies, and text. They believe it will work but have encountered some finicky issues.
Are there any improvements planned for image quality?
-Yes, they are working on improving image quality, particularly for small pixel artifacts. They believe they have found a way to significantly enhance it.
Is there a speed update in the works?
-There might be a small speed update that could make things 25-50% faster and cheaper. However, this update will be released after completing other updates.
What is the goal of the upcoming caption party?
-The goal of the caption party is to help teach the version 7 model the connection between images and language. If successful, it might be implemented as an official activity with rewards in the future.
What was mentioned about a new class of users?
-There was a mention of a new class of users who would be trusted with rating and captioning. Users might have to qualify for these rewards, which could potentially allow for larger rewards.
What is the current stance on video features and 3D models?
-The team is not super happy with the video features and probably won't see a version 6 model. However, they are optimistic about having a really good 3D model due to progress in hardware capture. The focus is on producing high-quality 3D rather than just exportable 3D.
What was said about the possibility of adding demographics to the feedback system?
-It was mentioned that they could add demographics to the feedback system in the future to find out who is really asking for each feature.
Outlines
📰 Mid-Journey Office Hours Recap for April 3rd
The recap begins with a recommendation for creatives to explore Medium, a website offering customizable prompts to save time. The main updates include a slower progress due to vacations, focus on website development with new social features, and initial testing with guides and mods. The launch will have limited social spaces with an aim to stress test the system. Personalization is being worked on, albeit at a slower pace due to multiple time zones. Style and random features are expected to return, with an algorithm in development to improve text accuracy and hands/ bodies. Efforts are also being made to enhance image quality and reduce pixel artifacts. A potential speed update is mentioned, contingent on other updates. A caption party is planned to improve the connection between images and language, with potential rewards in the future. A new class of trusted users for rating and captioning is being considered. Video improvements are still in the works, and while a V6 model is unlikely, a V7 model is anticipated with high-quality 3D models. The feedback leaderboard on the Mid-Journey website will receive more ideas for community rating. Demographics may be added to understand feature requests better. Lastly, consistent characters in generation may be possible in V7.
Mindmap
Keywords
💡Medium
💡Social Features
💡Personalization
💡Style Transfer
💡Algorithm
💡Image Quality
💡Caption Party
💡User Engagement
💡3D Modeling
💡Feedback Leaderboard
💡Consistent Characters
Highlights
Medium as a resource for creatives, offering customizable prompts and time-saving potential.
The progress on the website has been slower than usual due to vacations and staff being in multiple time zones.
New social features are being developed for the website, with initial testing involving a limited number of spaces and a larger number of participants.
Personalization is a focus, though it's advancing slower than desired due to the challenges of working across different time zones.
Style, random feature is expected to return, although the specifics are unclear and it may be related to dial tuning.
An algorithm is under development to improve the accuracy of hands, bodies, and text in images.
Despite receiving high feedback scores, occasional bad images are expected but will occur less frequently.
Efforts are being made to enhance image quality, specifically targeting small pixel artifacts.
A potential speed update is in the works, aiming to make processes 25-50% faster and more cost-effective.
The speed update release is contingent on completing other updates first.
An upcoming caption party aims to improve the version 7 model's understanding of the connection between images and language.
The possibility of an official activity where users can earn rewards for captioning and rating is being considered.
A new class of trusted users may be introduced for rating and captioning purposes.
Video features are still under consideration, but a version 6 model is unlikely.
Optimism for a high-quality 3D model in version 7 due to advancements in hardware capture.
Quality is prioritized over exportability for 3D models, though plans may change.
The feedback leaderboard on the Mid Journey website will receive more ideas periodically for community rating.
There are no plans to allow users to manipulate images with the Mid Journey model or to expand on not-safe-for-workplace features.
The potential addition of demographics to the feedback system to understand user preferences better.
Multiple consistent characters in a generation may be possible in version 7, not version 6.
An example prompt showcasing a serene double exposure image with stylization settings.
The speaker's social media handles for following their work on Instagram and Twitter.