Updates
Latest Updates
AllAnnouncementsChangelog
Date
4:47 PM
Announcement
Hi everyone, we have a few V8.1 related updates today:
- V8.1 is now available on Discord as well as midjourney.com
- We've improved sharpness and image quality for V8.1! This is most noticeable for SREFs and Moodboards but you should notice across all images (especially HD)
- To save compute during a server transition, we're making SD default for a bit. You can still switch to HD under settings panel or add
--hdafter your prompt. - The rating data you've giving us is continuing to help us improve quality for V8.2. If you want to keep helping us do ranking we've updated the images here: midjourney.com/rank-v8-1
Thank you and have fun! ❤️
Examples of the image quality improvements in crops taken from HD images
Date
3:02 PM
Announcement
Hi everyone!
We're preparing some major aesthetic updates for v8.1/8.2 and we want your help doing some image ranking to improve the model
Basically, we never did any ranking parties at full 2K resolution and in order to get the best possible HD pixels for everyone it would really help for us to rank some 'raw' images at the resolution from scratch
This will help tune the model to be more 'natively' HD/2K 'beautiful' and we think you'll all appreciate the difference soon
Rank images here. Tell us which one you think is more 'beautiful'.
Pro-tip: You can press the 1 key for the left image and 2 for the right image and 3 to skip.
Thanks again! ❤️
Date
4:33 PM
Announcement
Thanks for testing our V8.0 model over the last month! Today we're releasing our next version of V8 - V8.1
What’s new in V8.1?
- V8.1 has a consistent and familiar aesthetic in the spirit of V7. Moodboards and srefs are now super stable and will be everything you love.
- HD mode is now 3x faster and 3x cheaper. It’s so cheap we’re making it default for V8.1
- Standard resolution is now 50% faster and 25% cheaper. Standard resolution in V8.1 at full quality is as fast as V7 draft mode! It’s great for exploring and iterating *fast*. We’ve also added a button called “Run as HD” which will rerun any SD job as HD.
We’ve also brought back some community favorite features
- Image prompts and even image weights are now available in V8.1
- New Prompt Shortener, which will kick in anytime you go over the prompt length limits
- Updated Describe that gives longer more detailed prompts and matches the prompting style of V8
Please Note: Our V8 series of models are still only available on alpha.midjourney.com and are in an early testing phase. Things will change! Perhaps without notice! But good stuff is on the way.
It is likely that we will decommission our V8.0 model after V8.1 has been out for a few weeks. If you find something the old model did better than the new one please let us know in #ideas-and-features
What’s next? We’re going to try to give a quick shot for some V8 upscalers and then we’re moving onto our V8 edit, inpainting, and outpainting model upgrades.
Have fun everyone! And thank you for being a part of our community ❤️
Date
8:27 PM
Announcement
Hey everyone! We have some quick V8 Alpha related updates to make today
Relax mode is now available!
- V8 Relax is now up for all Standard/Pro/Mega subscribers
- Supports all commands (except using
--hdand--q 4at the same time) - There is a chance we run out of servers. If we do, we may change availability or speed of relax mode (until we get more servers online hopefully mid-late next week)
Two Versions of SREF / Moodboards now available
- We have a new version of SREF/Moodboards that is less accurate but is 4x faster/cheaper and also supports
--hdand works better with--pand--stylizeand--exp - This version is now 'default'
- You can use the old version by putting
--sv 6(style version 6) after your prompt - Our guides/mods are split between which is 'better' but certainly in my (David's) opinion, this new version is more 'fun' - so please enjoy!
Update on V8 Development
- We REALLY appreciate everyone's feedback so far!
- We are now working on a big training run for a new version of V8 (it might even be called V8.1)
- This new version will (probably) take 1-3 weeks to be ready
- When the next V8 comes out, we will likely remove the 'present' version of V8. So remember everything you're using now is Alpha and temporary.
- The next version we're working on is aiming to have improved default aesthetics, creativity, coherence, image prompts, better moodboards/srefs and possibly 2K default resolution
- Obviously we can't guarantee this but it's a big cool goal and worth our time to give it a shot!
Reminder: If you want to use MJ V8 go to alpha.midjourney.com and if you want to help us with ratings or feedback, check out Discord
Wish us luck and we hope you continue to have fun playing with the first early version of V8! <3
Date
5:08 PM
Announcement
Hi everyone! Today we’re letting the community test an early version of our V8 model on alpha.midjourney.com. We want your feedback and your help. Let’s explore what this new model can do.
V8 is an incredible model. It’s much better at following detailed directions and will still surprise you when you ask. Its ability to understand your aesthetics through personalization, style references (srefs) and moodboards is amazing. Images are more coherent and detailed. Text rendering works better than ever (when specified in “quotes”). And image generation is ~5x faster than before.
We’ve upgraded our web interfaces to support this speed. And more is coming. We have an improved conversation mode so you can just ‘talk’ in flow, a “Grid Mode” so you can focus on a single big set of images, and we’ve moved our settings into ‘sidebars’ so you can tweak settings without blocking your view.
V8 launches with support for multiple aspect ratios, --chaos, --weird, --exp, --raw. We support backwards compatibility with your V7 personalization profiles, moodboards, and srefs.
V8 brings a new --hd mode which natively renders images at 2K resolution and a --q 4 mode if you need a little bit of extra coherence.
Modes and Pricing: Presently, Relax is not yet supported. We are working on a new server cluster for Relax plus cheaper render modes. --hd, --q 4, sref, and Moodboard jobs are 4x as slow as regular jobs and each costs 4x (for now).
The best way to help us improve V8 is to rate images in the lightbox. Click an image in create then click “I like this image” or “I don't like this image” on as many of your V8 images as possible. You can use 1/2/3 hotkeys and arrow keys to rate images quickly.
Other Tips and Tricks: If you’re going for photos or a more ‘plain’ or ‘in control’ look we’d recommend switching to --raw right away or using moodboards or srefs to control your style. We’re still working on the ‘default’ aesthetics of V8, but we strongly encourage you to use your personalization and even crank it to –stylize 1000 asap. V8 is going to shine most right now when you rely heavily on our stylization systems and when you trend towards longer more specific prompting.
Please Note: This is an entirely new model with unique strengths and weaknesses, we want to learn from you what it's good and bad at but keep in mind it may require totally new prompting styles. So play around a bit!
It’s time to play! Show off what you’re making and let us know what you think on Discord. As the model becomes more mature, we’ll do a community-wide roadmap ranking session to help us figure out what to prioritize next.
Thanks again for everyone’s help with the V8 pre-release rating parties, and thank you so much for being a part of Midjourney! Let’s have fun out there and find new wonders on this vast and shared sea of imagination.
Date
3:13 PM
Change Log
Today we are removing the Rooms feature from the website.
Rooms was an experiment that taught us a lot about what our community needs and where our current infrastructure falls short. It tried to solve too many things at once, and didn't do any of it well enough. Rather than continue patching something that wasn't built to scale, we're making space to build these experiences properly.
We know this leaves some gaps, mainly around live support for our web users and a place for live events or collaboration. We’re actively working on filling those gaps with products focused on each use case. We'll share updates when we have them.
Moving forward, if you need support, you can:
- Visit our help page for immediate assistance
- Join #support or #promptcraft on Discord for live help
- Email support@midjourney.com
Thank you to everyone who shared something inside Rooms. We’ll bring them back in a better form.
Date
2:34 PM
Announcement
Personalization and Web Updates
Hi everyone, we've got a couple of small announcements today:
- We've got a new personalization interface on web
- This lets you make personalization profiles by clicking and scrolling through a list of images.
- This should be both faster, more accurate, and more fun
- This lets you also avoid you feeling anxious about 1v1 image comparisons (as before)
- We highly recommend everyone try personalizing Midjourney if you haven't already in the past!
- Go to midjourney.com/personalize to create a new personalization profile or continue building a personalization you started previously
We've added Personalization and Moodboards to niji 7 (our anime image model)
We're sunsetting rooms on web as we work on our next-generation social features.
- This should noticeably increase the general speed and performance across the entire web experience.
- We recommend people move to Discord for community events and live-chat.
Date
4:36 PM
Announcement
Hi everyone! We're now in the FINAL ROUND of the V8 image rating party.
This is the final round and means we're approaching V8 launch. This round will run all the way until final release.
The goal of this round is to calibrate our personalization systems specifically for V8. The images we show you in this round will be strongly different aesthetics and we want you to click the one you find most beautiful on a personal level.
Please go to midjourney.com/rank-v8 and help us rank images!
Directions:
- You’ll be shown two images at a time
- Select the one that you think is more beautiful (based on your own personal aesthetic taste)
- On a computer you can use the 1 and 2 keys to select images (1 is left, 2 is right)
- If you're not sure which is better, press "skip" (or the "3" key)
- These images are using random community prompts. If you see something NSFW press "Report Image" (Sorry!)
Please Note:
- These images are *NOT* representative of the new model we will release soon
- These ratings are very serious please try your best to provide us with good data.
Thank again for your help! ❤️ ❤️ ❤️
Date
3:14 PM
Announcement
Hi everyone! We’re now in ROUND 3 of the V8 image rating party!
The previous round was more focused on learning what's 'bad' and now round 3 is about learning about what's 'good'
Please go to the voting URL and help us rank images! midjourney.com/rank-v8
Directions:
Please Note:
- These images are NOT representative of the new model we will release soon
- These ratings are very serious please try your best to provide us with good data.
Thank again for your help! ❤️ ❤️ ❤️
Date
5:22 PM
Announcement
Hi everyone! We're putting the finishing touches on V8 and need your help!
Please visit midjourney.com/rank-v8 and help us rate images.
This rating party is focused on prompts that ask to generate text. We're going to do rankings on these for 1 day to see if we can tune the typography and text performance of V8 before release.
Directions:
Please Note:
- These images are NOT representative of the new model we will release soon
- These images are meant to be plain/boring/unopinionated and even bad to help our system learn what new things community wants
- If you're worried about quality of the images please don't! Images are always embarrassing when we do early rating parties. Getting everyone's help labeling 'junk' is a very important part of our process
- These ratings are very serious please try your best to provide us with good data.
Thank you so much for your help! We can't wait to share V8 with you soon! ❤️ ❤️ ❤️
Date
6:32 PM
Announcement
Hey everyone.
- We'll be doing 2 more rating parties this week and then will be releasing V8 shortly after
- We would still love your help rating more images in the Round 1 V8 rating party. Go to midjourney.com/rank-v8
- Your ratings are already helping the V8 model and we really appreciate your time and effort!
Thank you! ❤️
Date
8:59 PM
Announcement
Hi everyone! We're putting the finishing touches on V8 and need your help!
Please visit midjourney.com/rank-v8 and help us rate images.
Directions:
Please Note:
- These images are NOT representative of the new model we will release soon
- These images are meant to be plain/boring/unopinionated and even bad to help our system learn what new things community wants
- If you're worried about quality of the images please dont! Images are always embarrassing when we do the rating party number-one. Getting everyone's help labeling 'junk' is a very important part of our process
- These ratings are very serious please try your best to provide us with good data.
Thank you so much for your help! We can't wait to share V8 with you soon! ❤️ ❤️ ❤️
Date
11:41 AM
Change Log
We have made the following improvements and bug fixes:
- New aspect ratios: added 6:11, 4:5, 5:4, and 21:9 options to the ar slider
- Automatically strip irrelevant parameters (--cw/--ow) when cref/oref not present
- Automatically remove weird parameter from moodboard jobs, because it is not supported
- Improved drag and drop support for images into folders
- Updated the batch operations (download, like, trash, publish, and folder logic) to support 2000 items at a time
- Fixed folders not opening in lightbox view
- Fixed tablet UI layout issues
- Improved layout for mobile and tablets with long prompts
- Improved scroll anchoring and grid performance
- Addressed crashes caused by Google Translate browser extension
Date
6:20 PM
Announcement
Heya everyone,
Niji V7 is now live! For those who don't know this is our version of our image models specifically tuned for Asia and Anime.
Niji V7 features improved anime coherence, prompt understanding, text rendering, and sref performance.
We hope you enjoy!
For more information check out the blogpost here;
https://nijijourney.com/blog/niji-7
Date
7:23 PM
Change Log
Along with the Profiles launch today, we’ve updated the names of a few features to better match their functionality:
- The “Hide” filter has changed to “Trash”. Now, when you trash a creation, it won’t be visible on your “Create” or “Organize” pages, and it also won’t show up in the “Archive” tab of your profile. You can find your trashed creations in the “In Trash” filter on your Organize page. Others might still find them on the Explore page, in ranking, or see them if they have the job ID or creation URL. For more information, see our documentation about trash here!
- For Stealth users: What we previously called “Publish” and “Unpublish” is now called “Public” and “Private”.
We’ve also been improving our mobile experience over the last few weeks:
- Redesigned the image tray for image prompts
- Fixed view for create page with many parameters
- Lightbox swipe dismissal action
- Animate manually button bumps you back to prompt bar
- The mobile editor doesn’t glitch
- Organize page filters are scrollable
Date
2:04 PM
Announcement
We’re launching user profiles today. This gives you each an area on Midjourney where you can show off your images to others around the community.
Profile features
- You can customize your profile with a custom username, profile image, banner, short bio, and links to your socials
- You can spotlight your favorite images to curate what other people see
- You can follow other people (your followers are private and visible only to you)
- People can see your profile when they click your name on Explore or use your sharable profile link
For more check out our documentation:
Rewards for setting up profiles: To get the initial population of profiles seeded we’ll be rewarding 5 free fast hours to every person who sets up their complete profile (username, profile image, banner and spotlights at least 8 creations) within the next 24 hours.
Date
3:31 PM
Announcement
For the next three weekends - we're renting a big catamaran sailboat to take Midjourney members on a sunset cruise with musical performances and the opportunity to meet friends and members of the engineering team. Interested? Sign up here! midjourney.typeform.com/to/oo6kvC4Q
Date
2:36 PM
Announcement
We're doing a second high resolution style ranking party! This time, you'll rank only one image at a time to help us learn about detailed features. Please help improve our style system! midjourney.com/rank-styles-individual
Also, we're adding a new Midjourney TV feature to give you grids! Visit midjourney.tv/2 to do 2x2, midjourney.tv/3 for 3x3, or midjourney.tv/4 to view a total of 16 videos at once! Big numbers might probably crash your browser so be careful.
Date
3:31 PM
Announcement
Hi everyone!
We bumped up the number of styles in the style explorer and the style search feature by another 10x again today. That means we're up by nearly 200x since launch!
Don't forget you can search for styles in the upper right (video here)
We also have a style ranking party going on right now which will help us improve our style systems in the future: midjourney.com/rank-styles
More coming soon. Thanks! ❤️
Date
8:18 PM
Announcement
Heya everyone, we're doing a ranking party for styles!
We want your help to tell us which styles you find more beautiful.
By doing this we can develop better style generation algorithms, style recommendation algorithms, and maybe even style personalization.
Have fun! midjourney.com/rank-styles
PS: The bottom of every style has an --sref code and a ❤️ button. The top 1000 raters get 1 free fast hour a day. These rankings will not impact your personalization, but please take the ratings seriously.
Thanks for your help!
Date
7:47 PM
Announcement
Hi everyone! We're gonna start publishing educational videos on how to use Midjourney.
We're releasing our first ten videos today. Find them all on our youtube channel.
Let us know what kind of videos you'd like to see from us going forward, or share any feedback on the current ones on our discord. We'll keep at it.
Thanks! ❤️
Date
1:35 AM
Announcement
Heya everyone, a few quick updates!
- We increased the number of styles in the Style Explorer by 7x (including search)
- We fixed a major launch bug in the style hot/top feeds so now they are continually updating & evolving
- We've made it so when you look at 'Likes' you can filter them to be only images, videos or styles.
Have a nice weekend! ❤️
Date
7:11 PM
Announcement
Hi everyone! We're launching an early version of our style explorer today. It's like the video and image explore pages but now it's styles!
- Each thumbnail is a set of prompts rendered with a different SREF code (which sample randomly from the continuous latent space of all styles)
- You can save styles by pressing 'like'
- You can hover over styles and press 'Try Style" to render whatever is in your prompt bar (so you can just go through a ton of styles and render "dragon" for all of them, etc)
- You can also 'fuzzy search' through pre-rendered styles with the upper right search box on the style page. For example, if you type "photographic" or "anime" it will try to find the srefs that best correspond to those words
Try it out at midjourney.com/explore
There are more exciting updates coming for this feature and we'll try to add more styles continuously.
Have fun! ❤️
Date
5:47 PM
Announcement
New video options and a home for moodboards
Hi everyone, today we're releasing a bunch of small improvements to our services based on community feedback
HD video generation is now available for Standard Plan users
- Please be careful it is expensive
We're now letting you generate videos in smaller batches
- This means batches of 1 or 2 videos at a time
- Trigger this in the settings panel or with a command-line argument like.
--bs 1or--bs 2 - This will reduce costs by 2x and 4x accordingly for those who are extra cost-constrained. However, We highly recommend you keep it on 4 by default.
We've separated out Moodboards into their own page and made them accessible in the side menu
- Moodboards are the best way to set the style and mood for your projects.
- Highly varied moodboards will give you highly diverse images
- Narrow moodboards will give you focused images
We've changed the thumbnails on video jobs so they show the last frame
- This makes clips more easily distinguishable for traditional jobs but doesn't work well for looping or end-frame jobs.
- We will likely make another update to make the thumbnails based off middle-frames in the future. But for now we think this is better.
We've improved video moderation accuracy and you should now see fewer blocked jobs
Again, this is all based on community feedback, which we always appreciate.
Have a nice week! ❤️
Date
9:29 PM
Announcement
Heya everyone, we're releasing a HD mode for our video model today!
About HD Video Mode
- This HD mode is very expensive to run and we've increased the cost proportionally. It is ~3.2x more expensive than normal Midjourney video.
- HD mode is for professionals with higher budgets who don't mind paying the cost for the highest possible video quality
- HD mode is only available on Pro and Mega plans and does not work with relax mode
- HD mode will render at native 720p resolution. Like our normal SD resolution video model, we think our quality at each resolution is outstanding
- The higher resolution will reduce small scale pixel artifacts and may give you slightly better coherence or motion
- Many professionals report no longer wanting to use post-process upscalers when using HD mode.
Please note: We think if you're making videos just for fun, it is not necessary or desirable to use HD mode due to the additional cost.
How to enable HD Video mode
- You can enable HD mode on the website by clicking the settings button on the right side of the prompt bar then clicking "HD" under "Video Resolution"
We hope you have fun!
Date
5:07 PM
Announcement
Personalized feeds and midjourney.tv
Hi everyone! We're releasing two new exploration-related features today
Personalized feeds for images and video.
- There's a new feed on the explore page called "For You." This feed uses your V7 personalization to show images or videos made by the community that we think you'll like.
- This works for both video and images (just press image / video in the upper right)
Here are two direct links in case you want to go straight there right now: For You (Video), For You (Images)
Midjourney TV (Experimental)
- Midjourney TV is a experimental webpage and youtube stream which shows you random highly-ranked videos being made by the community
- It's available at midjourney.tv. The site will automatically select the videos that best match the aspect-ratio of your browser window. You can hover near the bottom of the video to see the username and prompt, and click to see the public job on explore
- We also have a youtube stream if you prefer--just be sure to set it to 1080p for the best quality: youtube.com/live/Jkbmn13F2hA
We hope you enjoy both of these fun features and find more beautiful things other people are making!
Date
8:33 PM
Announcement
Looping and End Frame for Video + Video in the Discord Bot
You can now create looping videos
You can now make videos with a specific start and end images
Video generation is now available on Discord
More information on how video works on Discord:
- To make videos click the "Animate" buttons under upscales
- If you have remix mode enabled, you can provide a prompt for the video
- To animate an image from outside of Midjourney, enter the image url at the start of the prompt and use the
--videoarg - To get a higher-quality .mp4, click one of the virtual upscale buttons on a video
- Videos can be extended. Turn on remix mode to change the prompt when extending
- To make a looping video, use
--loopin the prompt - To set an end frame, use
--end image_url
We’re excited to see what everyone makes. Have fun!
Date
3:19 PM
Announcement
Hi everyone,
We're starting to investigate opening up an Enterprise API for people to start integrating Midjourney into their companies and services.
If you'd like to apply, help us figure out what to provide, or just want follow-up updates please fill out our Enterprise API Application.
Thanks!
Date
5:28 PM
Announcement
Best ways to create more videos
Heya everyone! We're hearing a lot of people are running out of fast hours with our new video models.
So we want to give you some quick tips on how to get the best possible value with our current pricing scheme:
- You can buy extra fast hours individually, but upgrading to a higher membership tier gets you 2x more per dollar!
- Yearly subs get an extra 20% discount
- Pro and Mega tiers can help us test "video relax mode" right now (unlimited generations but with wait)
We hope you're having fun! It's amazing watching everyone explore ❤️
Date
1:32 AM
Announcement
Heya everybody! Glad you're enjoying creating and sharing videos. We're noticing that many Midjourney videos are getting "overly compressed" when uploaded to social media.
We've created a new optimized encoder which reduces this issue.
When saving an video to share, please right click your video and press "Save for Social Media". Or click the phone icon in the lightbox menu.
Date
12:35 PM
Announcement
Introducing Our V1 Video Model
Hi everyone!
As you know, our focus for the past few years has been images. What you might not know, is that we believe the inevitable destination of this technology are models capable of real-time open-world simulations.
What’s that? Basically; imagine an AI system that generates imagery in real-time. You can command it to move around in 3D space, the environments and characters also move, and you can interact with everything.
In order to do this, we need building blocks. We need visuals (our first image models). We need to make those images move (video models). We need to be able to move ourselves through space (3D models) and we need to be able to do this all fast (real-time models).
The next year involves building these pieces individually, releasing them, and then slowly, putting it all together into a single unified system. It might be expensive at first, but sooner than you’d think, it’s something everyone will be able to use.
So what about today? Today, we’re taking the next step forward. We’re releasing Version 1 of our Video Model to the entire community.
From a technical standpoint, this model is a stepping stone, but for now, we had to figure out what to actually concretely give to you.
Our goal is to give you something fun, easy, beautiful, and affordable so that everyone can explore. We think we’ve struck a solid balance. Though many of you will feel a need to upgrade at least one tier for more fast-minutes.
Today’s Video workflow will be called “Image-to-Video”. This means that you still make images in Midjourney, as normal, but now you can press “Animate” to make them move.
There’s an “automatic” animation setting which makes up a “motion prompt” for you and “just makes things move”. It’s very fun. Then there’s a “manual” animation button which lets you describe to the system how you want things to move and the scene to develop.
There is a “high motion” and “low motion” setting.
Low motion is better for ambient scenes where the camera stays mostly still and the subject moves either in a slow or deliberate fashion. The downside is sometimes you’ll actually get something that doesn’t move at all!
High motion is best for scenes where you want everything to move, both the subject and camera. The downside is all this motion can sometimes lead to wonky mistakes.
Pick what seems appropriate or try them both.
Once you have a video you like you can “extend” them - roughly 4 seconds at a time - four times total.
We are also letting you animate images uploaded from outside of Midjourney. Drag an image to the prompt bar and mark it as a “start frame”, then type a motion prompt to describe how you want it to move.
We ask that you please use these technologies responsibly. Properly utilized it’s not just fun, it can also be really useful, or even profound - to make old and new worlds suddenly alive.
The actual costs to produce these models and the prices we charge for them are challenging to predict. We’re going to do our best to give you access right now, and then over the next month as we watch everyone use the technology (or possibly entirely run out of servers) we’ll adjust everything to ensure that we’re operating a sustainable business.
For launch, we’re starting off web-only. We’ll be charging about 8x more for a video job than an image job and each job will produce four 5-second videos. Surprisingly, this means a video is about the same cost as an upscale! Or about “one image worth of cost” per second of video. This is amazing, surprising, and over 25 times cheaper than what the market has shipped before. It will only improve over time. Also we’ll be testing a video relax mode for “Pro” subscribers and higher.
We hope you enjoy this release. There’s more coming and we feel we’ve learned a lot in the process of building video models. Many of these learnings will come back to our image models in the coming weeks or months as well.
Thank you for being part of this journey with us - And have fun!
David
Date
12:46 AM
Announcement
Hi everyone! After months of testing V7 and the release of Omni-reference and the recent update to Style References for V7 - we're officially updating the default model to V7!
This means if you haven't tried V7 yet, your model will be automatically switched to V7.
If you wanna switch back to V6.1 go under the web settings on your prompt bar or type /settings on Discord and select V6.1 (or type --v 6.1 after your prompt)
What's new in V7 vs V6.1?
- V7 is a smarter model with higher image quality, better prompt understanding, and improved image coherence
- Omni-reference (
--oref) lets you put consistent characters and objects into scenes - Draft mode (
--draft) is 10x the speed of normal image generation - New sref and moodboard algorithms increase precision over V6 for defining the mood and style of your images
- Personalization profiles have been improved and are preferred by 85% of users
Have a nice day! ❤️
Date
3:48 PM
Announcement
Hi everyone! We're releasing our V7 Style Reference system today! It is now default and live for all V7 jobs.
Basic Instructions
- Drag a image into the style reference section on the prompt bar of the website. Or alternatively, type
--sref URL(with a url to a image) on Discord or web
More about sref V7 (style reference v7)
- It's much smarter at understanding the style of an image
- Much more likely to work even when the prompt differs strongly from the style
- Much less likely to get any undesired 'subject leakage' into your images
- There's a new
--sref random! the new styles should be much broader than before - If you want to use old sref codes you can either use V6, or you'll need to specify
--sv 4to fall back to the old model - The new sref also upgrades the functionality and accuracy of all moodboards
We hope you enjoy! Show off the new cool stuff you're making on Discord in #sref-showcase and let us know what you think in #ideas-and-features
Date
2:58 PM
Announcement
We're doing another pricing survey today...
Background: Our current plan is to launch a video model to as many people as possible. However, depending on many factors we cannot predict, this could be risky. We could start losing money in a bad way or we could run out of servers so severely that we may be under-provisioned for video for months.
Please help us fill out this survey to help us 'game out' what options we should consider if any of the above things happen. This will help us respond quickly to the situation as things develop post launch: midjourney.typeform.com/to/W0oi8erS
One last request: Please help us rate more real videos over the next few hours! We're close to the point where we close down voting and this is the last chance to help.
Please go there and pick which video you like better:
Thank you! ❤️
Date
10:13 PM
Announcement
Hi everybody. We're getting closer to a video model release, but we still need your help in two major ways:
- The video rankings data we've been getting has been a bit weird. Some people clearly like artifacts too much, so using the data is sometimes increasing the error rates in our video model.
As a compromise we're going to ask you to spend a little time rating "real actual videos". These are randomly sampled from around the internet. It's gonna feel weird!
Please go there and pick which video you like better:
This will help teach our model what combinations of motion and compositions people like seeing in video
- We also need your help figuring out how to price this release! We are extremely conflicted and may need to do multiple surveys to figure out how to have enough GPUs for everyone and how to not lose too much money
We did surveys before releasing our first image model and we still lost money our first month and had to make adjustments.
This first survey is also a little weird but it's focused on helping us figure out the optimal pricing that would make everyone happy.
Please help us fill it out!
midjourney.typeform.com/to/wuooDge6
Thank you so much for your help! ❤️
Date
12:22 PM
Announcement
Hi everyone! We've really appreciated your video ratings the last two days. Now we've made a second batch of video renders and we'd love your help rating them too.
These videos are even more wonky and messed up than the last bunch. It turns out that as we give the model more 'freedom' to do bad things, it also broadens its ability to learn the widest possible range of beauty from you ❤️
The directions and link are the same as before, but we'll re-write it all down for you just for extra clarity:
Please visit midjourney.com/rank-video and help us rate videos.
Directions:
- You’ll be shown two videos at a time
- Select the one that you like most (based on your own personal taste) and that also seem 'reasonably accurate'
- Please do NOT pick extremely broken videos even if you think they are funny
- On a computer you can use the 1 and 2 keys to select images (1 is left, 2 is right)
- If you're not sure which is better, press "skip" (or the "3" key)
- These videos are using random community prompts. If you see something NSFW press "Report" (Sorry!)
Please Note:
- These videos are NOT representative of the new model we will release soon
- These videos are meant to have a lot of low quality and broken stuff in it to give our algorithm as much freedom as possible in learning what people do and don't like
- If you're worried about quality of the videos please don't! It's always embarrassing when we do the first rating party. Getting everyone's help labeling 'junk' is a very important part of our process
- These ratings are very serious please try your best to provide us with good data.
If you think any of the videos are cool feel free to share on Discord in #rating-party
Thank you so much for your help! ❤️ ❤️ ❤️
Date
6:26 PM
Announcement
Hi everybody! We're putting the finishing touches on our V1 Video Model and we need your help!
Please visit https://www.midjourney.com/rank-video and help us rate videos.
Directions:
Please Note:
If you think any of the videos are cool, feel free to share on discord in the #rating-party channel.
Thank you so much for your help! We can't wait to share the V1 Video Model with you soon! ❤️ ❤️ ❤️
Date
7:03 PM
Announcement
Faster V7, ideas ranking party, and better moderation
Hi everyone! A few announcements today:
New feature ranking party!
Please help us rank what features should be on our roadmap and what's most important to you.
How does this work?
- Go to midjourney.com/ideas
- You’ll see 5 ideas and get 25 points
- Allocate the 25 points among the 5 ideas based on what you think to be most valuable (if it’s 2x more valuable to you, give it 2x more points)
- Do it a bunch of times (more helps!)
- Click leaderboard and see how the community’s votes are ranking all the features together! (it should converge in a few days)
Midjourney V7 image generation is now ~40% faster!
- Fast mode jobs render-time are reduced from 36 to 22 seconds
- Turbo job render-time are reduced from 13 seconds to 9
This is part of our ‘final optimization’ pass before making V7 default for the whole community and are necessary to prevent us from running out of GPUs.
As always, if you want to make your model ‘take more time’ try –q 2 and –q 4 (more updates will be coming in the future to these settings.
Note: Omni-reference jobs remain the same. Non-default settings may take longer (as before).
We’ve upgraded the AI moderation in the editor.
- We did an audit and found that blocked prompts were being properly blocked about 58% of the time.
- We trained a new model and the blocking accuracy is now up to 96%.
We hope you enjoy all these changes. Have a nice week! ❤️
Date
7:15 PM
Announcement
Hi everyone!
We're getting ready to make our V7 model the 'default' model but a lot of us still find ourselves 'enchanted' by earlier models. Maybe you really liked the colors of V3, the vibes of V4, the textures of V5, etc.
So we're doing a ranking party using images from all the versions of Midjourney!
We hope to use this data to see if we can get more 'old MJ magic' into new models, or maybe we'll even make some new style modes for V7 that are more inspired by the past.
Anyway, let's get started! Please go to this link and click the images you find more beautiful: midjourney.com/rank-old
Thanks for your help and have a nice weekend ❤️
Date
3:23 PM
Announcement
We're preparing to do a new community idea-ranking activity later this week. Before we put it up, we want to give everyone an opportunity to share ideas for what we should work on. These ideas may be included in the voting.
If you have time please fill out this survey to share your ideas https://midjourney.typeform.com/to/hHC2JVe8
Date
5:05 PM
Announcement
Hi everyone,
We're launching V7 model speedups today. This includes the release of the default fast mode, which makes V7 2x cheaper than before (now about the same price as a V6 job).
The model optimizations mean that jobs will take ~40 seconds on fast mode, and 18 seconds on turbo mode.
The new model should have slightly improved hands, but if you want the old version you can still get it via --q 2.
We've also added a new "experimental" --q 4 mode which might give you better coherence and details.
If you find any interesting differences or bugs with any of the versions please let us know in #ideas-and-features
Please Note: --oref jobs are still 2x the cost of a normal --fast mode job and draft mode jobs are unchanged.
Have a nice weekend! ❤️
Date
7:03 PM
Announcement
Hi everyone! Today we’re beginning testing of a new image reference system we call the Omni-Reference , this system can duplicate much of the previous functionality of ‘character references’ in V6 but it's also capable of much more.
What is Omni-Reference
- Omni-reference can be best thought of as a system of saying “ put THIS in my image”
- It can work for characters, objects, vehicles, or non-human creatures
How to Use Omni-Reference
- On web: Update your model version to V7 in the settings menu. Drag an image into the prompt bar and drop it into the bin that says 'Omni-reference'. There's a slider icon to control the strength
- On Discord: Type
--oref urlwhere the url directs to an image. Use the omni-weight parameter (--ow) to control the strength
About Omni-Reference Weighing
- There's a ‘omni-weight’ parameter called
–-ow(theres a slider in the web ui) which controls how strictly it adheres to your image reference. This parameter goes from 0 to 100 to 1000 (100 is default) - If you want to change the style of the image (such as photo to anime) you should lower the weight (ie:
-–ow 25) - If you want to make sure that a character's face is extremely visible (or that their clothes are preserved) you should try something higher like
–-ow 400 - Both
-–stylizeand-–expcompete for influence over the image with omni-reference. So if you have a high stylize or exp value, you must also use a higher corresponding omni-weight valuea person –stylize 1000 –-ow 400 --exp 100 –-oref person.png. - Please note: If you aren't using extremely high stylize and exp you should probably never want go over 'moderate' values of ow like
--ow 400or things may actually be worse
More about Omni-Reference
- Omni-reference should work with personalization, stylization, style references, and moodboards
- If you want a character to hold a sword in an image please make sure you say that in your prompt
a character holding a sword –-oref sword.png - If you want to style transfer a character and have a low omni-weight please make sure to over-specify the parts of your character you want to preserve
a anime woman with blonde hair and red suspenders --oref url –-ow 25 - It’s somewhat untested but if you have two characters (or object/character/etc) in your omni-reference image (either in the same image or just in two images side by side) and you refer to them both in your prompt can often get those two characters in your resulting image too
We know this is all a bit experimental. Honestly, there are SO MANY ways to use this feature. It’s hard for us to know what works well, what doesn’t work well, and what people want to use or improve most.
We need your help to test this! And your feedback to know what you want us to improve. Please give us your thoughts on Discord in #ideas-and-features and please show off what you’re doing along with the original reference images in #oref-showcase
Thanks everyone and have fun! ❤️
Date
6:59 PM
Announcement
Hi everyone, we're releasing a few quick updates:
New V7 image model update
- All images should now have slightly improved image quality, prompt accuracy, hand accuracy, and body coherence
- This applies to all images made with
--v 7. you don't need to change anything (you can even rerun old images with previous seeds and it may be consistent but better)
Improved lightbox editor interface
- When you click 'editor' inside the create feed we now have a new and improved editor interface based on community feedback
- This re-enables the image reel on the right and the default movement tool and brings smart segmentation to the lightbox editor all while keeping you close to your create feed
- The new editor ui now has vary and upscale buttons inside of it as well as easier access to various functionality
New --exp parameter
- This is a experimental image aesthetics parameter
--expgoes from 0 to 100, 0 is default- It is similar to
--stylize(and you can combine it with stylize), but it makes images more detailed, possibly more dynamic, creative, and more 'tone mapped' - The prompt accuracy will degrade as you go to higher values as well as the image diversity
- We recommend mainly using values of 5,10,25,50,100
- You will see a lot of change between 5-50 and not as much from 50 to 100
- As you go to extremely high values (above 25-50) you may notice it overwhelms --stylize and --p (we recommend lower values if you're mixing with other parameters)
We're getting ready to release omni-reference and fast-mode too but there are some small bugs in deployment that may delay things a bit. Stay tuned for further announcements!
Date
5:29 PM
Announcement
Hi everyone! Sharing some updates that we've made over the past two weeks.
Lots of editor improvements
- We've also opened up the external image editor to all membership tiers!
- We refreshed the full UI to make it simpler to edit and retexture images
- We combined editor experience for editing MJ images or external images. You can access the editor through the edit button in the sidebar, or via the 'Edit' button in the lightbox
- You can now import multiple images on layers. Use this to make creative collages of images or inventive new retextures
- Smart selection allows you to click on an object to be removed
- We've made the AI moderation smarter. You should see fewer prompts be blocked.
A new prompt bar for V7
- The new prompt bar now has separate toggleable buttons for Personalization, Draft Mode, Conversational (LLM) mode, and voice mode
- You can now use a AI to help you prompt or iterate over prompts when you click the 'conversational' mode button (the chat bubbles)
- Once you're in conversational mode you can also click a microphone to enable voice mode
- These modes are now separated from draft mode, but you can us them together too!
Improving draft mode
- We have improved the 'Enhance' button on draft jobs so that it will increase the quality of your base image without changing much in terms of details. This will be more subtle than a vary subtle, but will improve the quality of the base draft-mode image
- As a reminder, you can enable draft mode by clicking the ⚡️ button in the prompt bar, or adding
--draftto your prompt
Weird, tile, remix now supported for V7
- We've added
--weirdcommand back to V7. It mainly goes from 0 to 1000 but we're allowing 3000 like before. The main goal is to give you a command that lets you push things out of the normal aesthetic window into stuff that might be novel or weird in ways you might desire. Try using it with high values of--stylize!
Images made with --weird
- Use
--tilewith V7 to create seamless repeating patterns. Learn more about --tile - Remix mode is now available for V7. Remix is a bit of an advanced feature that gives you more control over variations. Find it by selecting "More Options" within the lightbox. Learn more about remix
A new way to help us improve V7
- Help us make V7 better by using the feedback options in the lightbox for whether you like the image, dislike the image or feel neutral about it
- Please take these ratings seriously! Community feedback like this helps make images better for everyone
- To make things fast we made it so you can also use 1,2,3 keys on your keyboard as well as arrow keys to flip through and quickly pick ratings
Thanks for being a part of our community! ❤️
Date
12:19 AM
Announcement
Hi everyone,
We're gonna let the community test an alpha-version of our V7 starting now
V7 is an amazing model. It’s much smarter with text prompts, image prompts look fantastic, image quality is noticeably higher with beautiful textures, and bodies, hands and objects of all kinds have significantly better coherence on all details.
V7 is the first model to have model personalization turned on by default. You must unlock your personalization to use it. This takes ~5 minutes. You can toggle it on/off at any time. We think personalization raises the bar for how well we can interpret what you want and what you find beautiful.
Our next flagship feature is “Draft Mode”. Draft mode is half the cost and renders images at 10 times the speed. It’s so fast that we change the prompt bar to a ‘conversational mode’ when you’re using it on web. Tell it to swap out a cat with an owl or make it night time and it will automatically manipulate the prompt and start a new job. Click ‘draft mode’ then the microphone button to enable ‘voice mode’ - where you can think out loud and let the images flow beneath you like liquid dreams.
If you want to run a draft job explicitly you may also use --draft after your prompt. This can be fun for permutations or --repeat and more.
We think Draft mode is the best way ever to iterate on ideas. If you like something click ‘enhance’ or ‘vary’ on the image and it will re-render it at full quality. Please note: Draft images are lower quality than standard mode - but the behavior and aesthetics are very consistent - so it’s a faithful way to iterate.
V7 launches in two modes: Turbo and Relax. Our standard speed mode needs more time to optimize and we hope to ship it soon. Remember: turbo jobs cost 2x more than a normal V6 job and draft jobs half as much.
Other features: Upscaling and editing and retexture will currently fall back to V6 models. We will update them in the future. Moodboards and SREF work and the performance will improve with subsequent updates.
Roadmap: Expect new features every week or two so for the next 60 days. The biggest incoming feature will be a new V7 character and object reference.
In the meantime - let's play! Show off what you’re making in #v7-showcase and let us know what you think in #ideas-and-feedback. As the model becomes more mature, we’ll do a community-wide roadmap ranking session to help us figure out what to prioritize next.
Thanks again for everyone’s help with the V7 pre-release rating party, and thank you so much for being a part of Midjourney. Have fun out there and find wonders on this vast and shared sea of imagination.
Please Note: This is an entirely new model with unique strengths and probably a few weaknesses, we want to learn from you what it's good and bad at but definitely keep in mind it may require different styles of prompting. So play around a bit.
Here's a peak at what's possible in draft mode
Date
3:18 PM
Announcement
We're now in ROUND 3 of the V7 image rating party.
This round will help us calibrate our aesthetic personalization systems.
We're gonna show you two strongly differing aesthetic pairs and we want your opinion for which one you personally find more beautiful.
Please visit https://www.midjourney.com/rank-v7 and help us rank images!
Directions:
Please Note:
If you think any of the images are cool feel free to share in #rating-party
Also, if you've never filled out our demographic survey we'd appreciate it! This helps us balance our default aesthetics across all the types of people in our community
Thank again for your help! ❤️ ❤️ ❤️
Date
6:49 PM
Announcement
Hi everyone! We're now in ROUND 2 of the V7 image rating party.
The previous round was more focused on learning whats 'bad' and now round 2 is about learning about whats 'good'
Please go to the voting URL and help us rank images! midjourney.com/rank-v7
Directions:
Please Note:
- These images are **NOT **representative of the new model we will release soon
- These ratings are very serious please try your best to provide us with good data.
If you think any of the images are cool feel free to share in #rating-party
Thank again for your help! ❤️ ❤️ ❤️
Date
9:58 PM
Announcement
Hi everyone! We're putting the finishing touches on V7 andneed your help!
Please visit midjourney.com/rank-v7 and help us rate images.
Directions:
Please Note:
- These images are NOT representative of the version we will release soon
- These images are plain/boring/unopinionated/bad to help our system ‘learn’ what new things community wants
- If you're worried about quality, try not to! Images are always embarrassing when we do the first rating party. Getting everyone's help labeling 'junk' is a very important part of our process
These ratings are very serious please try your best to provide us with good data.
If you think any of the images are cool feel free to share in #rating-party
Thank you so much for your help! We can't wait to share V7 with you soon! ❤️ ❤️ ❤️
Date
11:55 AM
Announcement
Hi everyone!
We've added a new folder and organization system to the website. Check out the video below for a nice overview.
Some highlights:
- A new Folders menu is available on both the Create and Organize pages
- When you’re viewing a folder, all the images you create will be added to that folder automatically
- Folders can be placed into groups to let you sort related projects or ideas.
- You can add images to multiple folders. If an image is within folders, they will be shown in the lightbox
- It’s now possible to download all the images in a folder just by right-clicking on it and choosing “Download”
- You can add items to folders by right-clicking an image, dragging images into folder, or clicking the folder button inside the lightbox
We hope these tools will let you manage more and bigger projects as you enjoy using Midjourney!
Date
8:08 PM
Announcement
Describe on Web + Moodboard Blending
You can now "describe" images on the web
If you can't find the right words to turn an image into a prompt, you can now use the describe feature on the web! To use it, right click on any image and select "Describe," or drag an image from your desktop to the describe zone. You'll see four text prompts generated from the image at the top of your Create page. To keep your creation feed clean, these prompts will disappear once you refresh the page.
Blend moodboards and style references
You're now able to mix together multiple moodboards as well as moodboards with style references. Just add multiple codes like "--sref 142710498 --profile drgmjoi 2jrqbw6"
Find all of your moodboards on the Personalize page. You can always mix moodboards and style references with your personalization profiles to tune it toward your preferences.
If you want to share moodboards with friends, you have two options. You can share a link to the moodboard to let people see the images and use the latest version as you make changes, or you can share a snapshot of the board with a code (e.g. --profile 2jrqbw6).
Hope you have fun!
Date
4:06 PM
Announcement
Hi everyone, we’re releasing an early version of our new model personalization infrastructure today
Flagship features
- You can now have multiple personalization profiles
- Setting up personalization is now much faster (up to 5x faster)
- You can now personalize models with moodboards of uploaded images
Try it out at midjourney.com/personalize
More Features / Information
- Personalization profiles both 'converge' more quickly and need fewer to get started. You'll need 40 ratings to get started, and the profile should be fairly stable by 200. You'll still notice that 2,000 ratings will improve the performance, but you shouldn't practically need 10,000 anymore.
- Moodboards take inspiration from the images that you add. As you add more diverse images to the board, the model will start to remix them in more complex ways.
- You can name model profiles and select one or many as default
- You can see all images made with particular profiles
We hope these features let everyone use personalization, whether you are working on multiple projects or with multiple people.
Here are some examples of what you can do with moodboards.
Have fun! <3
Date
4:48 PM
Announcement
Hi all! Today we’re launching a research preview of a new worldbuilding tool: Patchwork.
Patchwork is a collaborative, AI-supported infinite canvas for creating fictional worlds. It can help you go from vague story ideas to more complete ones; collaborate with other people to build a coherent setting together; and even assemble weird, collage-y visual stories from little scraps of images and text.
In the future, we plan for the characters, worlds, and other materials you create in Patchwork to be importable to other storytelling apps. This might let you bring your characters to life in interactive stories; gesturally act out story scenes between them to guide story text generation; revise story text in new LLM-based interfaces for creative writing; and more.
You can try Patchwork out at patchwork.midjourney.com. We’ve seeded the Patchwork multiverse with a few example worlds that might give you some inspiration for creations of your own.
Remember, this is a research preview, so there will be some rough edges as we roll it out! You'll need to have a Midjourney account, and to log in with Google. If you've always logged in with Discord, make sure to link your active Discord account with a Google account on your Manage Profile page.
To learn more about how to use Patchwork, read through the user guide and watch the video demonstrating the creation of a small world.
Patchwork Playthrough from Midjourney on Vimeo
Playing in picture-in-picture
Play
CC/subtitlesSettingsTranscript
QualityAuto
SpeedNormal
CC/subtitlesOff
Picture-in-PictureFullscreen
We can’t wait to see what you build! :)
Date
11:46 AM
Change Log
Hey everyone, we're doing a community contest for images made with the new editor. Please share your coolest examples of what you've made with the editor (retextures won't count) and we'll be giving out big fast hour rewards to everyone whose images we chose to show back off to public.
Submit your edited images here!
Thanks! ❤️
Date
8:29 PM
Announcement
Midjourney Idea Collection Fall 2024
Hi everyone,
We're doing a new survey this week to collect ideas that will help us plan our roadmap over the next 12 months Once we collect your ideas we'll throw them all together (along with our own ideas) into one big list and then put them up to vote to get a sense of what everyone would find most valuable.
You can do the survey here: https://tally.so/r/wkRJz1
We ask for ideas in a variety of ways but all idea prompts questions are optional and any ideas you give us are greatly appreciated
Thanks! ❤️
Date
6:05 PM
Announcement
Hi everyone, today we’re testing our new external image editor, image retexturing, and next-gen AI moderation systems.
The image editor lets you upload images from your computer and then expand, crop, repaint, add, or modify things in the scene.
We’re also releasing an ‘ image retexturing mode’ that estimates the shape of the scene and then ‘retextures it’ so that all of the lighting, materials, and surfaces are different.
All image editing actions can be controlled via text prompting and region selection. The editor is compatible with model personalization, style references, character references, and image prompting.
Along with these features, we’re testing a more nuanced and intelligent V2 AI moderator. This moderator holistically examines your prompts, images, painting masks, and output images. We think this may be the most intelligent AI moderator ever, and we hope some aspects of it make it out to our regular generation pipelines in the future so we can reduce false positives and give people more freedom than ever.
All of these things are very new, and we want to give the community and human moderation staff time to ease into it gently, so for this first release phase, we’re opening up these features to the following community groups:
- People who have generated at least 10,000 images
- People with yearly memberships
- People who have been monthly subscribers for the past 12 months (we tried to be more nuanced, but it’s complicated with the database, sorry!)
Known issues:
- If you ask something either extremely out of place or you ask for a tiny region to be changed, it might not give you what you ask
- Also if you put a tiny head in a scene and ask it to outpaint the body may be too big (so make the head bigger)
We hope you have fun testing these systems and watching the magic of Midjourney spill out into the world. We hope it inspires you to bring beauty to your surroundings, whether it’s remodeling your living room or imagining new kinds of fashion based on your personalized model and styles you love.
Learn more about how to use the editor in this video.
Midjourney Editor from Midjourney on Vimeo
Playing in picture-in-picture
Play
CC/subtitlesSettings
QualityAuto
SpeedNormal
CC/subtitlesOff
Fullscreen
We're so excited to see what you create!
Date
7:46 PM
Announcement
Hi everyone, personalization for the Niji (anime) model is available today!
If you rate at least 200 images on midjourney.com/rank-niji and then add --p after your --niji prompt, you can have personalized Niji model results. This personalization doesn't only change the style, but changes all elements of the images to suit your tastes. If you're a Niji model user you should try this out. We're sure you'll love it!
If you're not a Niji model user, you can personalize your Midjourney model at any time by rating at least 200 images at https://www.midjourney.com/rank and then adding --p after your prompt.
Don't be afraid to rank! It only takes about 5 minutes for 200 images.
Have fun! Learn more about Niji personalization on the the niji・blog
Date
12:52 AM
Announcement
We've opened up our web platform to everyone! Now you can explore and create in a custom-built experience. Check out an overview of how to use the site.
Welcome to Midjourney Web from Midjourney on Vimeo
Playing in picture-in-picture
Play
CC/subtitlesSettingsTranscript
QualityAuto
SpeedNormal
CC/subtitlesOff
Picture-in-PictureFullscreen
A few notes on signing in to the site:
- Sign in with your discord account on the website if you want to have history of your images made on discord
- You can merge discord and Google accounts under your 'account' tab if you haven't generated images under one of the accounts already.
Date
8:43 PM
Announcement
Hi everyone, we have a few updates for you today:
We're releasing a new web editor today!
This combines reframe / repaint / vary region / pan / zoom into a single unified interface. We think this makes editing your MJ images way more seamless than before and is a huge step forward.
Here's a video showing how the editor works
Room Syncing
We're now syncing images and messages between Rooms on the web and channels on the Discord server. This means that messages and images in certain rooms/channels are reflected back onto Discord and vice-versa.
Channels and Rooms with message mirroring are:
- #daily-theme <-> Daily Theme
- #prompt-craft <-> Prompt Craft
- #general-1 <-> General Chaos
- #superusers-create <-> Superusers (for people with at least 1000 images)
Opening Up Creation on the Web
Everyone who has made at least 10 images on Midjourney can create on the web. Go to midjourney.com/imagine to start creating!
Date
1:46 PM
Announcement
Hi everyone, we’re releasing V6.1 of our Midjourney Image Model today!
What’s new in V6.1?
- More coherent images (arms, legs, hands, bodies, plants, animals, etc)
- Much better image quality (reduced pixel artifacts, enhanced textures, skin, etc)
- More precise, detailed, and correct small image features (eyes, small faces, far away hands, etc)
- New 2x upscalers with much better image / texture quality
- Roughly 25% faster for standard image jobs
- Improved text accuracy (when drawing words via “quotations” in prompts)
- A new personalization model with improved nuance, surprise, and accuracy
- Personalization code versioning (use any personalization code from old jobs to use the personalization model and data from that job)
- A new
--q 2mode which takes 25% longer to (sometimes) add more texture at the cost of reduced image coherence - Things should look “generally more beautiful” across the board
Other important notes
- V6.1 currently does not have a new inpainting/outpainting model so if you use zoom/reframe/repaint/vary region it will fall back to the V6.0 model
- If you want to change back to V6 just go to /settings and click V6 or type --v 6 after your job (or click the settings panel on the website)
Future plans: We will likely release a V6.2 model in the next month or so with further improvements, but first, we need to gather as much data as possible of people using V6.1. So as of today we’re making V6.1 the default model on Midjourney for all users.
This release has been a long-time in the works and has been guided by the priorities submitted and ranked by the community at midjourney.com/ideas
We thank everyone for their involvement in this process and we hope everyone enjoys the new release! Have fun! ❤️
Date
1:44 AM
Change Log
We've got a list of assorted changes and improvements going live today:
Creation:
- Fixed and improved feedback when uploading images into the imagine and chat inputs.
- Dragging images into the imagine bar now works properly on Firefox.
- Notification appears on the create tab when creating an image from the lightbox with an action button.
- Improved error design and messaging in creation.
- Fixed issue where Basic Plan users could select `Relax` mode, causing an error on every creation.
Help:
- We're tying together the various help resources into a single /help page, which adds a help chat w/ bot. We'll continue evolving this page over time to help new and experienced users alike.
- Loading the system status report is now significantly faster.
- Bug reports are now sent to a dedicated QA team for review.
Rooms:
- Rooms are now publicly visible to anyone with an account, though only those with web creation access can participate.
- Opening large rooms from the lobby should be much faster.
- There are new default rooms for Prompt Craft, Newbies, and Help to help users learn.
General:
- Option to follow your system's light/dark mode preference.
- Many small stability fixes.
Date
6:50 PM
Announcement
Hi everyone,
We're starting a new research project to explore the origins of beauty... Basically, when do we find something collectively beautiful, and when do we each see beauty differently?
We're doing this by comparing the personalized models people are making to the answers in several new surveys.
The first survey is about demographics, how do things like age, gender, country, etc effect what we find beautiful?
Take the survey at midjourney.com/surveys .
It takes ~5 minutes and everyone who completes it gets 2 fast gpu-hours as reward
We'll try to use this data to share deep new insights about human aesthetics and potentially even build new features to let you explore those insights yourself! All data is private, you can request deletion at any time on your account page, and you can skip any question you want (you'll get rewards for whatever percentage you answer).
The next survey will be more about personality and should come in a week or two.
Thank you so much for volunteering your time to help us push the boundaries of science. We look forward to sharing the results with you and exploring the meaning of it all together ❤️
Date
3:05 PM
Announcement
Hi all,
We're opening up orders for Midjourney Year One — an archival book that showcases our community's explorations and Midjourney's development over the past year. It's a beautiful & physical snapshot of the last year as well as a fusion of the many issues of the official Midjourney Magazine that you might have missed. Order a copy at mag.midjourney.com
We have also made a few back issues of the magazine available to complete your set. These can be found at mag.midjourney.com/midjourneymag
Date
5:50 AM
Change Log
Temporary Disabling of Rerun Button for 'Submitting' Jobs
#hotfix
We are temporarily disabling the Rerun button for jobs in the Submitting phase. We are working on solutions and hope to re-enable this feature within the next week.
Currently, clicking the rerun button for jobs in this phase results in an error every time. While this issue is not new, a recent change has unintentionally increased its occurrence.
Date
9:53 PM
Change Log
Hi all! We've made a few changes to the website in order to make it simpler to use. Some changes of note:
- 'Pan' and 'Zoom' tools are now unified and called 'Reframe'
- 'Remix' is now found within 'Reframe'. This allows you to change the prompt when panning or zooming.
- 'Vary Region' is now called 'Repaint'
- Error messages are more descriptive and don't require hovering to get more information
- The sidebar is simpler. 'Rank Images' and 'Rank Ideas' can now be found under the 'Tasks' page
- 'Report Bug' can now be found under the 'Help' menu
Date
9:37 PM
Announcement
Hey everyone! We're releasing an early test version of model personalization today. Here's how it works.
What is Model Personalization?
- Every time you write a prompt there's a lot that remains 'unspoken'. Our algorithms usually fill in the blank with their own 'preferences', which are really the combined biases and preferences of our community
- But of course everyone is different! Model personalization learns what you like so that it's more likely to fill in the blanks with your tastes
Model Personalization Requirements
- Right now model personalization learns from votes in pair ranking and images that you like from the explore page
- You need to have roughly 200 pair rankings / likes in order for the feature to work
- You can see how many ratings you have on the pair rankings page, or by typing
/infoon Discord
How do I use Model Personalization?
- Just type
--pafter your prompt, or turn on personalization for all prompts using the settings button in the prompt bar - When you enable personalization you get a 'code' added after your prompts, you can share this code to so others can use the personalization effect being applied to that image
- You can control the strength of the personalization effect by using
--s 100(0 is off and 1000 is maximum and 100 is default)
PLEASE NOTE: The personalization is not a stable feature right now, it will change as you do more pair ratings (subtly) and we may be pushing out algorithm updates over the next few weeks. So just assume that it's a 'fun feature constantly in flux'.
This is a totally new frontier of image synthesis. We hope we discover a lot in the process of doing this.
Have fun!
Prompt of "a home --ar 1:1 --p"
Date
8:31 PM
Announcement
Create Rooms to Use Midjourney with Friends
We're now opening up the ability to create rooms for all Midjourney members with at least 100 images.
How does it work?
- From the rooms tab, click "Create room" and give your room a name, description, and password.
- Share the room link and password with friends so they can create with you.
Now vs future:
- Right now we only let people make private rooms (which are password protected rooms for your friends / workplace).
- In the future we will let people make public rooms with no password protection and a global gallery.
We hope you all enjoy creating images together!
Date
8:30 PM
Announcement
We're opening an early test of our "creation rooms" on the web today for all users who have made more than 1,000 images with Midjourney. We'd love your help testing! To use them, you'll need to visit our testing website at https://alpha.midjourney.com/
A few cool features in rooms:
- It has text chat and voice chat
- Text is separated from images so creation doesn't affect chat
- You can make images together, see only your own, or view images that the room has liked (filter buttons on top)
- You can easily drag images into chat to talk about them
- You can resize the chat / image columns by hovering then dragging on the space in-between them
Please Note: If you create images in stealth in a room other people can see them (just like in Discord)
Some high level plans right now around rooms as context:
- We plan to make it so anyone can create both public and private creation rooms on the website
- There will be a lobby of public rooms and the ability to directly link to private rooms
- We might create more rooms if one gets too crowded / frenetic
- After watching a small number of rooms for a bit we'll open up creating rooms on your own
We hope you have fun testing the rooms. For now it's DESKTOP WEB ONLY (mobile is a secondary priority for us and will come later)
Room plans (hopefully next week): We're going to add in AI noise-cancellation for voice chat, then we're going to try adding a few more rooms and then we're going to open up rooms to more people.
Longer term: We'll let people make private rooms, then we'll let people make public rooms
We've made a few more updates in the past few days as well.
We now provide a easy 'sref code' for --sref random
Basically, if you type --sref random it'll replace it in your prompt with something like --sref 123456 (with a random number) and then if you type that random number back in like --sref 123456 you'll be able to get back to that style again
We've released a new algorithm that adds a quantitative measure of "controversy" to the community idea leaderboard
Check it out here: https://www.midjourney.com/ideas?type=leaderboard
Basically, "100 controversy" is average, 50 means it's half as controversial as average, and 200 means it's twice as controversial as average. We're defining "Controversy" as the difference in importance between "enthusiasts" and "skeptics."
Have fun this weekend!
Date
9:07 PM
Announcement
Hi everyone! We're testing new algorithms for "exploring the latent space of visual styles" and are opening up a simple and fun tool for everyone play with.
You can now type --sref random after your prompts and get a totally random style.
For example, here are 121 separate runs of /imagine a sailboat --sref random
If you end up liking something you find, use --sref url with the url to the image to get more of it
This is meant to be a test of a underlying algorithm that we make more advanced tools for in the future.
For now we hope you just have fun!
Date
10:05 PM
Change Log
#alpha#feature
We've replaced the link to the bug-reports channel with a new tool. Now, you can detail your issue, include relevant prompts or jobs, and attach screenshots. Additionally, we can now collect essential background information to aid in investigating and reproducing bugs.
The reporting tool also has a quick view of https://status.midjourney.com/, allowing you to check on any ongoing issues before lodging a report.
Date
9:35 PM
Change Log
Fix for --sv not submitting in web
#alpha
- Fixed small incompatibility that meant
--svwouldn't be submitted with prompts on the web.
Date
7:42 PM
Announcement
New Style Refs and the return of free GPU hours
Hi everyone, two quick announcements:
- We're releasing 2 new algorithms for style references plus a new 'versioning' feature
- The latest version of the style reference algorithm (now default) is more precise and works for both MJ V6 and Niji V6
- You can set the version of your style reference algorithm with
--sv - There are 4 options to chose from:
--sv 1The original style reference algorithm we released with--sv 2The one that was default up until today--sv 3A new version of sv1 with better performance and more 'vibey' than sv2--sv 4A new version of sv2 that with improved performance (this is default)
- We are turning back on free GPU hours for image rating.
- The top 2000 raters every day get 1 free fast GPU hour
- Do image ranking here: https://www.midjourney.com/rank
- These ratings help improve our algorithms, populate the explore page, and also will be used for upcoming model personalization features
- Ranking goes fastest if you use the 1/2 hotkeys on PC or use it on mobile
- Please take rankings seriously. We've deployed a new bot detection & abuse systems. Using bots for ranking or mashing random buttons over and over may result in your account being banned. Don't do it!
Hope everyone enjoys the new releases. Thanks!
Date
10:00 PM
Change Log
#feature
- Added the
/updatespage to stay appraised of the latest announcements, gallery changes and Office Hours directly on the website. - Added simple control over which category of updates you want to see in the main feed and receive pings about.
Date
1:59 AM
Announcement
Rolling back Niji --sref update
We found a major bug with the new —sref for niji so we've rolled that back to the old version. The new sref is still available for normal v6.
Date
8:00 PM
Announcement
Hi everyone, we're releasing v2 of the "style reference" feature today.
It should be much more precise about understanding style and it should be much better at not 'leaking' non-style stuff into your images.
We've also improved the way the weight parameter --sw works too, so that it's more meaningful and also has aa bigger range.
What are style references again?
Style references give you the ability to reference the style of another image or to keep a consistent style across many images. It's similar to image prompts but focuses solely on the style.
How do I use it again?
To use a style reference, type --sref URL after your prompt with a url to an image. To increase the influence of the style, apply a style weight (--sw) that is closer to 1000. To diminish the power of the style, set the weight closer to --sw 0 (default is 100). You may add multiple style references as well.
Here's a quick silly example using a crayon drawing as a reference. Anyway we hope you have fun! ❤️
Date
8:00 PM
Announcement
Hey @everyone @here we need to halt our free-gpu rewards for image-ranking until further notice due to extensive automation / abuse.
Image ranking powers our front-page and helps us improve our algorithms, but a large group seems to be abusing it for free hours (perhaps as much as 50%) and it's starting to hurt our systems.
So basically, we're moving from a reward system based to a system based on altruism and fun to see what happens.
If you want to do ranking to find new prompts or images, or to help improve our algorithms, or help populate the frontpage, go ahead, we love you for it! https://www.midjourney.com/rank.
If you're looking for free hours, we're not doing it anymore, since it seems to be incentivizing bad behaviour.
We hope this helps us get better data, improve our models and make the explore page amazing.
Our team is also working on new special features like model personalization that will use your rating data, so your hard work will hopefully pay off in the form of new abilities.
Thanks for understanding and for being a part of this amazing community ❤️
Date
8:00 PM
Announcement
Introducing Character References
Hey everyone we're testing a new "Character Reference" feature today This is similar to the "Style Reference" feature, except instead of matching a reference style it tries to make the character match a "Character Reference" image.
How it works
- Type
--cref URLafter your prompt with a URL to an image of a character - You can use
--cwto modify reference 'strength' from 100 to 0 - strength 100 (
--cw 100) is default and uses the face, hair, and clothes - At strength 0 (
--cw 0) it'll just focus on face (good for changing outfits / hair etc)
What it's meant for
- This feature works best when using characters made from Midjourney images. It's not designed for real people / photos (and will likely distort them as regular image prompts do)
- Cref works similarly to regular image prompts except it 'focuses' on the character traits
- The precision of this technique is limited, it won't copy exact dimples / freckles / or tshirt logos.
- Cref works for both Niji and normal MJ models and also can be combined with
--sref
Advanced Features
- You can use more than one URL to blend the information /characters from multiple images like this
--cref URL1 URL2(this is similar to multiple image or style prompts)
How does it work on the web alpha?
- Drag or paste an image into the imagine bar, it now has three icons. selecting these sets whether it is an image prompt, a style reference, or a character reference.
Shift+Clickan option to use an image for multiple categories.
Remember, while MJ V6 is in alpha this and other features may change suddenly, but V6 official beta is coming soon.
We'd love everyone's thoughts in #ideas-and-features.
We hope you enjoy this early release and hope it helps you play with building stories and worlds
Date
7:00 PM
Announcement
Hey everyone we're launching V6 Turbo tonight!
What is it?
- Turbo mode jobs go 3.5X faster at 2x the cost (~10 second vs ~35 seconds)
- These run on the fastest possible GPU machines (that's why it's both faster / more expensive)
- It's great for groups, public events, whenever you're in a rush or have the 'need for speed'
How to use it / how does it work
- Type
/turboand all your/imaginejobs will go 3.5X faster (at 2X fast-minute gpu cost) - Turbo mode currently supports regular
/imagine``varyandremixjobs (but not sref yet) - Any jobs not supported as turbo jobs run as normal jobs (at regular cost)
It's pretty wild do have V6 at 3.5x the speed. Enjoy and have fun! ❤️