Latest Midjourney News

Midjourney Office Hours

Dec. 13, 2023

Website
– the next thing coming on the website is collections this week, downloader next week
– image generation for 10k+ users is now
– folders and collections are coming
– a new Explore page with more diversity and novelty

v6
– majority of people working on it
– internal rating party is done
– starting to test the tuning back in from the rating party
– First external rating party possible Fri/Sat this week
– Second rating party after the first one
– all current features will not be available at release e.g. inpainting
– Prompting will be different! No more 4k, 8k, trending on art station, ultra HD, masterpiece (don’t do that says David) – use natural language
– release: after 2nd rating party if all looks good – v6 alpha will be released

Midjourney Office Hours

Dec. 6, 2023

Main things working on: Website and v6

Website
– the next thing coming on the website is collections this week, downloader next week
– image generation for 10k+ users: next week
– a new Explore page will drop in December on the website

v6
– massaging final image quality now
– prompting will be different – prompting strategy needs to change e.g. cleaner prompts, less junk words
– will have better image prompting
– most likely will keep a 2×2 grid and 1024 resolution)
– better upscaler (faithful and creative upscaler)
– all current features will not be available at release e.g. inpainting
– internal rating party next week
– community rating party – for all MJ members (after the internal one)
– doesn’t look like v6 will be released next week – possible 3rd or 4th week of December

There will be a Niji 6.

Other
After v6
– will start to train video models (this will not be like other video products out there)
– will work on character and style consistency
– will look at faster and slower models

The current (old) style tuner will not work on v6.

Midjourney Office Hours

Nov. 22, 2023

Priorities: Website and v6

Website
– drag and drop download selector and collections – next week or so
– image generation mode, will be first be available for users who have generated 10k+ images – next week or so

v6
– the first version is trained – now in tuning phase testing internally
– much more coherent, understands language
– playing with subtle setting e.g. how well understand language
– deciding how big the model will be
– text capability broke – will consider adding text capability into v6 anyway
– may not have all features right from the start (haven’t decided yet)
– bigger jump than from v4 to v5
– hoping to have it ready for release in December (before the holidays)

Style tuner most likely will not work in v6. Would create a new tuner for v6.

v5.3 is cancelled.

There will be a Niji 6.

Other
After v6 release will start to work in character consistency.

Midjourney Office Hours

Nov. 15, 2023

Website
– working on a downloader and collections
– image generation mode, when it’s released (1-2 weeks), will first be available for users who have generated 10k+ images
– will have a better explore page on the website soon

v5.3
– won’t be released – going straight to v6

v6
– the main thing the team is working on (along with the website)
– test version the team is using internally
– better image quality
– better text understanding, very likely v6 will write text (may have two versions, one with text, one without)
– two weeks of internal training (release candidate)
– not going to rush it out before tuning it
– hoping to have it ready for release in December (before the holidays – if all goes well)

There will be a Niji 6.

Other
3D is limited with data.

Midjourney Office Hours

Nov. 8, 2023

Website
Will push out small updates, possibly even today
Image generation on the website will roll out in stages, first, the 10k+ users will get access

v5.3
Potentially a v5.3 release (80-90% chance)
Could be similar jump from v5 to v5.1
Could have a coherency boost
Possibly next week

v6
Most people are working on it
Test versions are currently rendering text
Bigger jump than from v4 to v5
Earliest at the end of the month (could be later)

Niji
Working on a style tuner for Niji
There will be Niji 6 after the v6 launch

Other
Good results with video and 3D but limited with data

Have a better upscaler, but it’s slow, so not sure what to do with that.

Midjourney Office Hours

Nov. 1, 2023

For the current v5.2 model
– will try to release style tuner today (doing final checks now)
– How it works: type: /tune and then enter a prompt, then choose options > choose the ones you like
– it will cost GPU hours to use the style tuner

Website
– will push some small updates this week (for organization, search, etc.)
– working on the next phase (where you can generate the images)
– power users will have access to old alpha site to test web image creation (10k plus users)
– new explore page: 1-2 weeks

v5.3
– may do some a v5.3 before v6, but not sure yet

v6
– priority with the website
– no estimate release date at this time
– Niji update with v6

Most users don’t get to 1000 images.

Midjourney Office Hours

Oct. 25, 2023

Recently released

Upscaler 2x and 4x
Niji app (Niji Journey)
New website (phase 1 – in beta)

Website
– focusing on building an environment that helps people be in a flow state
– will move from beta to active site
– next: image generation capabilities

v6
– considering moving from feature-based release cycles to date-based release cycles
– training two different types of models; will be able to start getting results from them this week (then decide whether to combine them, use only one, etc.)
– working on a fairly conservative v6 to start with
– may not have text in the first round of v6, maybe in v6.1
– a more accurate understanding of a short(re) prompt (1-2 sentences vs. a really long prompt)

The goal of Midjourney is to have fun and to explore.

Style personalization
– quiz, tuning, style quiz? Still not sure what to exactly call it
– will ship it soon (tomorrow?)

Number 1 priority after v6 and web release: character consistency.

Midjourney Office Hours

Oct. 18, 2023

v5.2 Upscaler
– subtle mode
– 2x and 4x (up to 4k x 4k)
– more details, raises the quality of images
–  will come out really soon (this week)

v5.2 Aesthetic and style tuning
– will include new aesthetic and style customizer/tuning
– testing it, not sure what to call it (personalization test?)
– trying to make it better
– conflicted whether to release it now as is or continue to try to make it better
– may release it this week

The Niji Team has made a Niji Journey for Android and iOS.

Website
– Phase 1 will be released next week
– Phase 2 sometime next month

v6
– better prompt understanding, coherence, better knowledge
– possible release towards the end of November

Will release Niji 6 after v6.

Midjourney Office Hours

Oct. 12, 2023

TL;DR Upscaler release is estimated for the end of this week.

Recap.

Website
– Phase 1 will be released first: can view and search images, better organization
– Phase 2 later with image generation and social features

Version 5 upcoming features:
Upscaler:
– more faithful/subtle and more creative – more subtle works and is ready to go
– can go pretty high resolution, may start with 2x
– can update upscalers later
– may release basic upscaler by the end of this week

Personalization test:
– testing now
– not quite ready to be released

v6
– working on it, individual components are working (individually); now need to put them together
– possible release towards the end of November (hard to estimate dates; it’s running behind)

Other:
3D
– making progress

Video
– have nice video datasets now

Midjourney Office Hours

Oct. 4, 2023

Upcoming releases

In about 2 weeks: Phase 1 of the website
– fast and easy to find and view your images
– no image generation (will be in phase 2)

In about 2+ weeks: v5.3 Upscaler and style personalization
– both are in the testing and tuning phase (aka close to ready)
– upscale to 2k x 2k images, possibly higher
– you will be able to share some of the personalized styles

In 6+ weeks: v6
– huge jump from v5 – really different with different capabilities
– if dragging, may release v5.5

Other updates:

3D – continue to have better 3D – may come out this year (not necessarily for game development at first but for fun)

Video – looking better – maybe the end of this year, early next year

3D and video will be built on top of v6 and v7.

Midjourney Office Hours

Sept. 27, 2023

• aesthetic personality tests – help customize your image making
• upscaler in the works (2k x 2k)
• estimated release in October

Playing around with color e.g. HEX code release
Possible aesthetic improvement (would/could call it v5.5 – if released)

v6

• better image quality, better language understanding, better blends, better resolution
• bigger batch size (maybe 2x or 4x)
• 4-8 weeks? (unlikely to come up in October)

Website

• approaching Phase 1 release of the website
• better lightbox, better design (new codebase)
• phase 1 will not have image generation

Other:

– may ask users to describe pictures in detail for fast hours

– working on 3D and video, will likely release 3D first (before video)

Midjourney Office Hours

Sept. 20, 2023

Updates to v5 (maybe v5.3?)

• personalization – will help you make what you want without having to use all kinds of settings
• upscaler – testing in progress: would double the resolution
• general aesthetics improvements are coming, and more coherent
• will be released relatively soon; no release date

v6

• much bigger jump than from v4 to v5
• better image quality, better text prompting
• will be able to prompt TEXT
• no release date at this time

Website

• will launch in two phases
• Phase 1: ability to view, search etc. of your own images
• Phase 2: will have image creation and some social features
• no release date for Phase 1

3D

• no 3D in v6 (between v6 and v7)
• should come in the next 6 months, much further along now than before
• 3D model generation, not sure if mesh quality will be that high

Other:

New /describe at some point. Not sure when it will ship.

A new Niji style is in the works.

Playing with controlnet type things. When to push controllability knobs or how to make things a lot better.
e.g. color palette

Improved AI prompt moderator and will keep updating it.

Midjourney Office Hours

Sept. 13, 2023

v5.3
(not sure if these releases will be called v5.3 or something else)

• updated aesthetic system, makes the images a bit more plain but more accurate to the prompt, more coherent
• a personalized aesthetic system (personality test)
• the goal of the personality test is to give you your preferred aesthetic – you can adjust it
• no specific release dates at this time

Looking at new upscaler technologies

• may be able to ship an upscaler for v5

v6

• much bigger jump in performance than from v4 to v5
• most of the team is working on this
• lower resolution to start, upscale to e.g. 2048 x 2048
• pretty much better everything
• no release date at this time

Website

• Website will roll out in two main phases
• Phase 1 will roll out relatively soon but won’t have ability to generate images
• Phase 1 will basically replace the current website with much better search capability of your own images and viewing them
• Phase 2: a combination of image creation and some social features enabled
• Phase 3: additional things like tools to help you find e.g. styles you like etc.
• no release date for Phase 1

Other

• working on 3D and video
• excited about recent 3D results
• early concepts for video in the works

Midjourney Office Hours

Sept. 6, 2023

Inpainting

• can generate better images- when inpainting the full canvas
• may release this separately soon, for example, as –alt

v5.3

• working on the release
• new aesthetic system that may allow people to personalize things
• thinking about, e.g., asking a brief quiz to find out your preferences before generating images
• estimated release: in September

v6

• early version ready, but it will take a while to get to the release version
• should also be able to generate text, better prompt understanding
• no estimated release date at this time

Website

• working on a big web release
• will be released in two parts
• the first release will have a new code base (no image generation turned on at first)
• you can look at, search, and download your images
• a much better way to search your images and a lightbox
• some things will be removed: no followers (maybe one day – not like some other platforms)

Mobile

Native mobile stuff in development

Niji

The team is starting to work on the next Niji Style.Other

• recently updated the moderation system
• auto-appeal is turned on
• video and 3D prototypes going – looking better
• eventually will have video and 3D at the same time

Midjourney Office Hours

Aug 23, 2023

Inpainting

• Called Vary (region)
• may ship some more inpainting features

v5.3

• may release the style improvement to prompt accuracy (would/could be called v5.3)
• no release date or certainty if this will happen

v5.5

• entertaining the idea (may or may not happen)
• would be slight improvements vs. a big one with v6

v6

• priority
• will take some time to cook
• more controllability, more knowledge, better text understanding, more styles, more diversity, ability to generate text, better resolution – a lot!
• a lot of energy and work going to v6
• no release date at this time

Website

• priority
• making progress
• super high-quality UI and UX
• adding a lot of features
• no release date at this time (but when it drops it’s going to be really, really, really good)

Describe

• working on better /describe, more accurate
• will ship it outside of v6

Other

• working on 3-D, may get it this year (need data)
• working on video – may come together with 3-D
• will have an upscaler at some point
• will have Niji 6

Midjourney Office Hours

Aug 16, 2023

v5.3

• trying to release inpainting this week
• an aesthetic update possible but may be pushed back a bit (slightly more accurate to text) – if the aesthetic update is included will call it v5.3
• will be a pop-up within Discord

v6

• still behind schedule, could release something more conservative e.g., call it v5.5, but would like to really bring a bigger update (with bigger changes)
• will take some time to cook
• focusing on making it much, much better at understanding text and prompts
• no release date at this time (a few more weeks)

Website

• working hard on making the core experiences great
• no release date at this time

Demographic Research

• biggest aesthetic differences between ages (young vs. older) and male vs female
• the current system has a bit of an older male bias, e.g. when using a vague prompt (not specifying a gender)
• working with the knowledge from the surveys to make the system better

Other

Working on infrastructures and databases to keep everything going.

Midjourney Office Hours

Aug 9, 2023

v5.3

• possible release next week
• inpainting and aesthetic – in Discord
• may be only available on desktop

v6

• working on it, adding more performance
• will be more coherence, better language understanding
• the grid size may change, but not sure yet (internally, they use bigger grids e.g. 4×4)
• no estimated release date at this point

Website

• a lot of people working on frontend of the website
• will be both for desktop and mobile
• launch date is getting pretty close, but no specific date at this time
• Explore will be much better
• Create page will be much better

Surveys

Interesting results from data
– 80% are male users
– people who are rating images are older users
– male and older user biases in the ratings
– interesting differences, e.g., male vs female and older vs younger aesthetics

For example:
Females – prefer more of a v3 type look/aesthetic, males – high contrast aesthetic

Geographic differences
– e.g., Japanese vs. Europeans have different aesthetic preferences
– difference even between, e.g., Germany and UK

MJ has a feedback loop where it learns a bit from the community, so it can have users’ biases. They have started to look at this and start detangling things.

They will publish results and stats later.

Other

Sometimes this month may test a 2x speed for the pro and mega tiers.

Storytelling research continues.

Working on video. No specific timeline.

Secret projects on-going (as usual).

No API anytime soon.