mjofficehours

[mjofficehours]

2023-02-15

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

During the pre-hours sirkut (naysmith) mentioned that they made this prompting guide about working with color palettes and I think it's rad so here's a link https://docs.google.com/.../1G9w5trNqG2uasybrzR1MxLO.../edit

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific. You can attend the Office Hours on the MJ Discord server in the voice channel called "Main Stage". They are not recorded for later listening.

You're welcome.

1. Working on v5. Lots of issues in v4 that can’t fix so pushing fwd to v5. Sharper, higher resolution, hands better (but not perfect), more realistic.

Tiling looks great in v5.

Might start with no upscaler and all images at 1024X1024 and possibly add upscale rlater..

2. In-painting isn’t working great out-painting is working good. Not sure when it will be out.

3. Real time drawing - they have proof of concept but don’t know when it will ship.

4. New rating system test soon for v4, hopefully will test it this week.

5. Working on linking daily theme contests with new rankings system. Prizes!

6. Still working on free tier.

7. He reiterated their new approach of making specific versions for specific features/advances. Rather than trying to mash all these things into a single update. This is why they are breaking v5 up into v5, v6, and v7.

Currently releasing a new version every 4 months but would like to get v6 and v7 released quicker.

8. Hard to tell what aesthetics will be like for v5. They will do that at the end.

9. They are pushing hard on the website.

People with 10k plus image accounts have access to BETA of new website's UI. (I'm not sure I understood this correctly. I have well over 10k images and didn't get an invite).

10. New MJ website search feature starting to take shape. Hope to get it working in the next month. Hope to have feature for searching all images available next week. This feature will add moderating challenges though, so they are being mindful of that.

11. The Discord server is getting too big. Scary. May have to break it up. Will probably make a trial server just for trial users.

12. Image to text is promising but a ways off.

13. 3D will happen before motion. Motion solutions that look good are very expensive so MJ is going to avoid for a while.

14. In the future image grids will be larger - meaning that instead of getting 4 images per prompt you will get many more.

15. Next year this time UI will be unrecognizable compared to now.

16. Someday there will probably be specific hardware (probably a chip) made specifically for these kinds of system..

17. AI text models have already run out of text to train on on the internet

18. Someone asked David what some of his favorite prompts are and he mentioned some that he was currently playing with. The image for today's Take-away is based on one of those prompts that he listed.

[mjofficehours]

2023-02-08

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

I've been doing these for 2 months now so I have some perspective about how goals and practices seem to have evolved in that time and I'll add a brief note about those observations at the end.

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific. You can find listen to them on the MJ Discord server in the voice channel called "Main Stage".

You're welcome.

1. Wider aspect ratios are available 1:1 - 2:1 and everything in between like 16:9.

2. Every time there is an update it will change the aesthetics to some degree. The more specific your prompts the less they will be affected by small changes in the aesthetic style.

3. NIJI is getting an update soon to fix bugs

4. At this point team is leaning towards conserving energy and pushing towards the v5 model. v4 is not as sharp as it should be and they can’t figure out why but v5 doesn’t have that problem so it makes sense to focus on v5.

Initial images for v5 look really good but they don't know yet how well it will hold up or what fixes might be necessary. They expect testing to be 3-6 weeks.

The images coming out of v5 look so much better and have such higher resolution that he's confident that most people will be happy with them without upscaling.

They are planning to include upscaling but it gets very expensive once the images are already that high rez. He thinks upscaling may only be needed in the case of printing large images. He did not get specific about DPI or anything like that but he seemed very excited about v5 image quality.

v5 will not be the version that improves text generation (text within an image like a title or a logo) we will possibly see that in v6.

v5 will have much better hand generation. It still won’t be perfect but you should notice a huge difference. They expect hands and bodies to improve with every version.

5. Better language stuff (in terms of MJ understanding your text prompt) will probably come in v6. A better knowledge base overall will probably be v7.

Focus now is in turning out one major feature or improvement per update instead of trying cram a bunch of features into one massive update.

6. Working on a new rating system. The launch party for v5 will be an online rating party. You will all be invited whenever it happens.

7. They are continuing to work out how to adjust payment tiers. he mentioned a lot of idea but it seemed very stream of consciousness. His main goals are to improve the $10.00 tier and to somehow create a free tier that doesn't bankrupt the company.

He also mentioned that the only reason they can afford the GPUs to train v5 is because of people paying at the $30.00 and $60.00 tiers so he's very thankful that so many are doing so, and making it possible for him to do this work.

8. Magazine is coming. Don’t worry they won’t randomly send it to you. You have to ask for it. First one is free. You will get a DM asking if you want it once it is ready to ship.

9. They are thinking about planning a souped up daily theme. He did not elaborate but I like soup.

10. They are getting a lot more AI people hired "which is great". They are playing with some real -time stuff (artist's tools). Inpainting and outpainting are still very much on the board.

11. Still working on search functions but it will be fast once we get it.

12. Still need LOTS more front-end people. Send an email to careers @ Midjourney.com

13. He is thinking about an API for individuals but not one designed for corporate professional use.

14. "AI images are like flowers and the users are like bees the flowers are trying to attract with their beauty"

Here's my take-away from 2 months worth of take-aways.

A. The goal posts are more concrete and (to me) seem more realistic. back in December he was talking about having a new update every month and expressed optimism that v5 would be released before the end of December.

I'm actually comforted by these more conservative projections for when updates will happen and what they will contain. It bodes better for the company as a whole imo. Things take time.

2. There is a commitment to customer service and building the community and the main obstacle to improving those areas is the sheer scale involved. David mentioned today that they are specifically choosing not to market MJ because he doesn't think they could handle the influx.

They really want to provide a free tier, community resources, and a better user experience overall but the fact that there are over 10 million of us complicates how to do that.

[mjofficehours]

2023-02-01

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific.

Apparently they are publishing office hours notes in Discord now which is lovely but I found them a bit difficult to parse so I will keep producing these for a bit longer but I'm optimistic that documentation will continue to improve and then I'll rest from my labors.

You're welcome.

1. Wider aspect ratios coming soon 2:1, 1:2, 16:9, 9:16 should work much better when it ships. Better composition, less stretching and fewer borders. Going to try to ship it asap. Should be ready. Takes a lot of coordination to implement changes like this so it may be a few days before we see it. He hopes we really like it.

2. They have a good idea of a way to improve the word filter. Hope to ship it out in the next couple weeks

3. v5 looking good but needs testing with full system. Hitting first gate towards release beginning of next week. 2-6 weeks worth of testing after that.

Higher res, sharper, faster, better details, hoping image quality boost is significantly high enough to make people happy. Don’t know how close we are but progress is happening

4. David mentioned v6 several times for the first time I remember.The gist of it was that v5 is going to focus on quality of images and stability of the platform rather than many of the new features that they originally envisioned and v6 will probably be similar. They feel that the most important features of image quality (and other important features such as wider aspect ratios) should be the priority for now. So things like animation and 3D are still in the plans but are probably a ways off.

5. Still working to refine text filter. Refining image filter prompts is still proving difficult with no clear path forward.

6. Might do one more style update for v4 and the style will definitely be different in v5.

7. Still working on search features for website. Working on building social features into search engine to highlight unusual images.

8. They did lots of hiring. If you have a proven track record and completed projects on your resume for coding front end, AI, or video streaming utilities please DM one of the MJ mods on Discord (not FB).

9. Real time tools for artists. Draw and the image swirls around like a magic paint brush. This kind of real time image creation with AI becomes a kind of video streaming. (Not in the sense that your work would be public) it’s what would allow you to change an image in real time via MJ. They don't have anyone on their team with expertise in streaming so if you are a rock star at coding video streaming and have the resume to show it please reach out to one of the MJ mods on Discord (not FB)

10. They are building web and mobile team but it’s lagging due to needing more staff.

11. Still trying to find a way to get free tiers for people without cash. Rating images in exchange for GPU.

12. There’s progress with in-painting and out-painting but it's still not working as well as they want it to.

13. You weren't imagining it. Relax mode gets slower the more you use it until eventually your images always go to the end of the line. This is why your images may show up faster during relax mode if you are prompting when less people are on.

They do this because otherwise the company would go bankrupt providing GPUs to the small percentage of the most active users that make upwards of 100,000 images per month and that are already costing them $1,000.00 a month even with gating relax mode.

14. Gate keeping specific text prompts is silly because they aren't going to be relevant a year from now. Use them and share them now because they will be obsolete this time next year.

[mjofficehours]

2023-01-25

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

HOLD ONTO YOUR HATS PEOPLE WE HAVE DOCUMENTATION! Very exciting news today there is a new channel in the official MJ Discord server called whats-new that has detailed updates of progress being made. AND there is an official Docs landing page with a lot of very useful information about using MJ collected in one place and nicely organized.

I will copy/pasta the most recent update after my write up but I recommend checking out the channel yourself. Thank you so much to whoever worked on providing this much needed documentation

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific.

You're welcome.

1. Image prompts have been improved and Blend has been improved.

2. MJ hit 1M users on at once today. Getting close to 10M accounts

3. Check out the new mini polls on the MJ Discord server.

4. They need more front-end people. Please apply.

5. Search is getting better but database size is challenging

6. Getting closer to higher aspect ratios in v4. May debut in NIJI first.

7. Team is experimenting with inpainting and outpainting. Results are promising but not ready yet.

8. v5 is training right now. Wont be released until late February if then. Will release in phases rather then waiting to have everything at once.

9. Priorities for v5 are sharper images, higher res, trying to get higher language abilities in there (understanding prompts).

10. Working on ideas for tweaking word filtering. Image filtering is more difficult to improve.

11. Web features are coming.

12. Still working on free tier. In providing a free tier the goal is to lose money in a way that benefits the people who are paying for the service.

13. More vague but intriguing talk about a “magical paintbrush”

14. V5 and v6 will both have increased GPU costs. And training them is very expensive. Most of the money is going towards the GPU cost of training v5.

15. Magazine is in the works and swag later

First update from the snazzy new "whats-new" channel on the MJ Discord posted by fnuckle(Community Liaison Witch)

"So, what’s new in 2023 so far?

- Prompts that use “soft banned” words should no longer disappear when generating in users DM’s - Changed the initial job message from saying “0%” to "waiting to start"

- Fixed the --stop parameter when being used with --niji

- Instantly deletes discord message of jobs when it gets canceled in users DM’s (to cancel a job in discord, react to the message with the emoticon)

- Fixed "report job" hanging when clicked (to report a generation from discord, right click a message and go to apps > report job)

- Added --niji suffix as the default for /blend on the niji bot

- Improved Image Prompting

- The two main things it improved is the blurriness, and the “white-shifting” problem. Using image prompts should now generate sharper and better contrasted results.

- /blend command has been updated

- Added optional “dimensions” field with “portrait, landscape, square” options. Portrait mode is 2:3 and Landscape is 3:2.

- Removed the --v 4 default suffix and added support for user-set suffixes (users normal suffixes will now automatically appear on /blend generations)

- The “dimensions” field will take precedence to any --ar command set in the user’s suffixes

And lastly, my most favorite new update~ We have deployed our new official docs!! Go take a look - extra (fake) points for you if you find the page full of cats" https://docs.midjourney.com/

[mjofficehours]

2023-01-18

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific and there is no other way to get these updates.

You're welcome.

Lots of people are working on the website and mobile app. Hope to have it soon.

Website search should show big improvement soon also.

They are looking for ways to make User stats for how weird/original/unusual images are in order to promote those rather than tending to feature the most common tropes as they do now.

V5 models are training now. There was some discussion about people wanting to donate images and David took it seriously enough to make a micro poll about it on the MJ Discord so you can head over there and vote if you have feelings about it.

They have not given up the dream of improved aspect ratios in v4 but aren't sure when, or even if, it will happen.

There are a couple people working on tiles. They are making progress.

They are working out some prototypes for in-panting and out-painting. He said it was going well and he seemed optimistic about it being a feature at some point.

They are working on the UI, Backend structure, and optimizing GPU.

They will be experimenting more with free tiers soon

"There is a non -trivial chance we will focus on v5 rather than trying to fix v4." v5 release will not be in January might be next month. Might not.

Still thinking about tools for people who draw. "Magic paintbrushes". He expressed that he felt the best way he could serve the technically skilled artistic community is to give them the tools they need to thrive in this new environment.

Moving towards image prompts. He reminded everyone to check out the new /blend command for uploading and mixing images

Peter Morbacher (sp?) is going to provide a digital painting tutorial in MJ Discord. It will not be recorded.Tuesday Jan 24th at noon Pac. It is listed in events on the Discord server.

Future of art will be weird (in a good way) because you won't have to appeal to the masses with a generic product. Artists will be world building and making movies and comics. He thinks artists will be more in demand than ever

17,000 away from 9 million users

It’s not stability vs growth it’s transformation vs decay

[mjofficehours]

2023-01-11

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

EDIT --q2 in v4 is still broken but it doesn't eat GPU. It doesn't do anything.

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific and there is no other way to get these updates.

You're welcome.

They are working actively on better wide aspect ratios in v4

v 5 is working good in some areas and not in others. It has sharper images and fewer artifacts. Prompting is better but not great yet

They are working on changing the ranking and rating process. Improving both the interface and rewards.

They need a better feedback system. They are going to experiment with a combination of surveys and AI. For now please respond to the micro-surveys in the official MJ Discord server.

No word on when v5 will ship. Maybe this month. It’s definitely not as good as it could be so it’s worth waiting.

They have launched some new payment tiers. See the official Discord server for details.

New command soon /blend that will allow users to merge images more efficiently. The UI allows you to attach images directly instead of having to post it to discord, get the url etc…

Actively working on search function for website. It’s going well and once it's finished should be robust enough to handle the massive number of images

Trying to make billing and subscriptions easier to navigate

Focusing on features for people who like to draw. The future involves crafts and crafts people. It’s a priority because it adds to the overall robustness of the community.

3D is being put aside temporarily to focus on these more complicated skill oriented features and overall improvement of quality to 2D images. But 3D is definitely coming anyways within the next year or two. 3D requires different approaches depending on what is being created whether it's object creation, character creation or environment creation. Right now he’s most interested in character and environment creation.

They use the best environmentally-green cloud server they could find. It uses solar and water power to provide GPUs.

He feels like the aesthetic range of work coming out of MJ is not as wide as he would like. Things are going to get a lot more weird before things get a lot more specific. (He didn’t elaborate on exactly what that entails).

The goal for the UI is to make it feel like an extension of your mind not a conversation with an exterior servant

[mjofficehours]

2023-01-04

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

I don't think he mentioned this today but the default version for MJ is now v4 so you can stop typing --v 4 at the end of prompts if you are doing that. Also q2 in v4 is broken right now. All it does is eat GPU.

Disclaimer: I am not associated with Midjourney, Discord, or this FB group in any official capacity. I do this because many people can't attend the weekly Office Hours via Discord on Wednesdays at noon Pacific and there is no other way to get these updates.

You're welcome.

1. Improvements to upscalers and higher/different ratios to v4 are not performing as well as hoped.

2. Its a balancing act trying to make improvements to v4 while pushing ahead with v5. Especially since v5 solves a lot of those issues in v4 that need improving. So they are shifting towards v5 but still want to improve aesthetics of v4.

3. v5 is progressing nicely and they are still trying to get all the features they want in it rather than rushing to get it out. He is reluctant to mention the features because he doesn't want to disappoint anyone if a feature doesn't make it into v5.

4. They are focusing on certain improvement in v5. More details. Better accuracy and resolution. Better response to prompts.

Read text better (I'm not clear if this referred to prompts or text placed in images).

5. More advanced tools for building longer series of images.

Some tools will require you to be more of an artist.

Important to have a version for people who like to draw. 90% of users won’t get good images (on the drawing MJ interface) but the 10% who like to draw and want to do that for a living will be able to create original work faster and have more fun doing it.

It's important to provide the drawing UI soon so people can see there’s a craft to it. Even though it only serves a small portion of users.

Side note: there was a lot more to this topic but the gist is that he understands now that there are people who like to draw and want to do that for a living. Not everyone who makes images is interested in telling a story. Many of us just like drawing and so giving us an interface that allows us to use our drawing skills to get the results we want is going to go a long ways to reaching out to the artist community.

As someone who likes to draw I found this very encouraging and I can't wait to see what this UI looks like.

6. They are still working on image creation sites that are separate from Discord. Mobile app looks like it will be 10X better than Discord (on mobile).

7. They are looking at increasing the value of the $10.00 tier and creating a free tier where you pay for your sub by helping them with data collection (I think that means ranking images).

8. They might like to expand service to China (the way they have to Japan with NIJI) . He also noted that China is the first country to regulate AI images (they require watermarks).

EDIT: this was a passing comment as part of a broader discussion about what govt regulation might look like in the future and should not be read into too deeply.

9. GPU situation is much better. Things should be good unless the user base grows much faster but for now there is plenty of padding.

10. They are planning some contests as a way to hire new staff. They want to see what people can make. That's all the details I got but it sounded like they will tell us more soon.

11. they are thinking a lot about how to improve filters. Word filtering should improve over the next month

12. He doesn't expect to have a truly stable version until v8 or v10

13. They are still trying to get search working for the MJ website. He’s sorry everyone. It turns out it’s really complicated at this scale.

14. They have no plans to incorporate a chat bot directly into MJ but if they did make one it would be way weirder than what's currently out there.

[mjofficehours]

2022-12-21

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

I don't think he mentioned this today but the default version for MJ is now V4 so you can stop typing --v 4 at the end of prompts if you are doing that.

1. Wider aspect ratios (16:9, 9:16, 3:2, 2:3) for V4 are expected to be here soon

2. He would like to have better upscalers for V4 but no one is working on it right now.

EDIT: to clarify he expressed that it's a very important feature. I got the impression that he has been thinking a lot about the need for good resolution for quality prints but it's complicated and he's still trying to figure it out.

3. Goal for resolution is 1024X1024 default

4. Tiling is technically possible as a feature in MJ but there is no one to work on it yet

5. New web-based MJ image creation site might get pushed out to 5k users for testing soon

6. Would like to make mobile app that works independant of website or Discord. This is a relatively high priority project because he want people to be able to use their phones for MJ more easily.

7. Niji journey just launched.

8. Stuff coming in January

A. Tools for artists who can draw (he was vague about this but sounded very excited.) As an artist who draws I'm thrilled to hear this and am looking forward to finding out what the heck it is.

B. Better results for people who don't think of themselves as artists (my phrasing) using better ways to determine personal style preferences (the personal style buckets he discussed last week).

9. GPU Situation should be good now

10. There is something called "Prompt Battles" in the voice chat on the MJ Discord server now. I'm sorry I didn't understand how it works but people are enjoying it and it sounds like it's worth checking out.

11. A caller asked for more voice chat channels. These are going to happen and will have different purposes. There will be at least one devoted to MJ help and tutorials.

12. They want to do 3D but we aren’t there yet

13. Even though they plan to expand web/mobile interface they have no plan to move it completely away from discord because community is important.

14. They would like to be able to provide an Image detection function that would allow you to hover over an image and know what stylistic eras/influences are in it. This is possible but we aren't there yet.

15. The future is both wonderful and terrifying and filled with water and tigers and it's important to be able to tell them apart so you can work with the water and avoid the tigers. (again my paraphrasing - hopefully I'm close enough to the gist of well over 60 minutes worth of transcendent metaphors).

[mjofficehours]

2022-12-14

This week's take-aways from the Office Hours Discord session. I'm sure I forgot/missed some. Please help fill in the details if you can. TY.

1. Server issues are a big concern and they are doing all they can. Part of the issue is a shortage of GPUs all across the server hosting platform they rent from (possibly due to holiday pressures). In any event he's actively looking for ways to increase the available GPUs to 10X average MJ user base rather than the 2X it is now. It's going to get a lot better soon but probably not instantaneously.

2. the --q 2 command should be disabled. It's not currently helping anything and it's making things slower for no benefit. They are probably going to shut it off. (I personally recommend turning it off. I did and it definitely sped up my results with no loss in quality)

3. There is a back button on rankings now

4. They will soon make v 4 the default version instead of v 3.

5. v 5 should be out sometimes this January

6. He expects people using image prompts (prompting from another image or combining images) to become half the user base or more after v5 drops. He mentioned that the people most comfortable using text prompts tend to skew older and that younger people tend to think more visually.

7. They are considering offering artists micro-grants to help struggling artists buy supplies as a way of directly supporting that community.

8. They are hashing out the details for a free tier where you do work doing ranking for them (as you can now with the $10.00 tier). They don't know yet how much work it will entail.

9. Someone asked David why he made MJ and he gave this lovely nuanced and detailed answer which I can't possibly remember all of but it boiled down to a desire to encourage people to envision the solutions needed to survive the 2030s by using MJ for self reflection, as a tool for imagining solutions, and coordination of new communities based on shared visions.

[mjofficehours]

2022-12-07

Did anyone else listen in on the "Office Hours" discord audio session yesterday? These were my take-aways.

1. David Holz is much more patient than I am.

2. Their focus is on providing the best software for lay-people. They are prioritizing features that make it easy for anyone to create images for fun. Rather than features that make it convenient for professional to manage their businesses (like multi-seat licensing).

3. They have no plans on adding more prompt codes (like -- q 2) and plan to move away from them entirely. We won't lose those options it will just be easier for people to select them.

4. They are aware that images skew darker and under-saturated. They are working to address this bias.

5. They are aware that human images skew towards attractive young white women and are working to address that bias.

6. They expect to see 100% improvements in image resolution, details, and results within 6 months, possibly sooner.

7. They are working to make it so that people can choose a preference for a whole set of parameters which they called "putting people into buckets" meaning that you would choose a particular "bucket" with a pre-set aesthetic starting point that you can then adjust from. You could change buckets at any time.

8. They expect to be able to add limited movement to images within the year so you can make .gifs. Eventually they expect people to be able to make whole movies and in the more distant future (10 years) entire 3D custom virtual environments and interactive assets.

9. They are keenly aware that any changes make 80% of their users happier and 20% sadder and they would like to fix that but don't know how.

10. They are aware that the banned words are often restrictive but don't want to make the mods life hell in a world where they are moderating millions of users.

11. MJ users generate 10 images of cats per second.