AI Mindset · Model Cheatsheets
Midjourney

The Art Director Who Thinks in Images

Midjourney’s terms don’t just gate commercial use at $1,000,000 in revenue. They gate ownership. A company over that line, sitting on Basic or Standard, doesn’t own what it makes: you “must be subscribed to a ‘Pro’ or ‘Mega’ plan to own Your Assets.” Put that in front of legal before the first upload. Past that, V8.2 has been the default since July 24, 2026, and as of October 15, 2026 there’s no V8.3 and no V9 anywhere. A Thinking rerun is in alpha testing with no published cost. HD is a toggle. There’s no API, and the terms ban automated access outright. Four tiers, $10 to $120 a month.

Verified October 15, 2026Ownership gated at $1M revenueV8.2 default since July 24V8.2 HD documented at 2048 pxNo API, and automation is bannedHD is a toggle, not a defaultThinking Mode in alpha, cost unpublished
1 / Meet Midjourney

A fast visual collaborator, not a universal assistant

Think of it as an image department with a strong point of view. Give it a clear creative job. Generate enough breadth to make a real choice. Then use references and editing to develop the route you picked.

Personality

Aesthetic, divergent and fast

It throws out visually persuasive options quickly. That makes it excellent for early exploration and dangerous when a team mistakes polish for strategic fit. Midjourney calls the V8.2 look more creative, bold, sophisticated, edgy and fresh, which raises the need for curation rather than lowering it.

  • Strong default visual taste
  • Rapid iteration
  • Best with active human curation
Deploy it for

Seeing possibilities before committing

Campaign territories, product visualization, editorial illustration, environments, style exploration and pre-production frames.

  • Explore 20 routes, not three
  • Settle the direction with pictures, not memos
  • Spend production budget on selected ideas
Choose it when

The decision is visual

Reach for Midjourney when words stop being enough to judge the idea. The team needs concrete options in front of it.

  • Taste and mood matter
  • Consistency matters across assets
  • A human will own final selection

The professional Midjourney loop

The useful unit of work isn’t one prompt. It’s a managed creative cycle.

Brief
Define the visual job

Name the audience, the asset, the communication goal, the space you need and the constraints. Not just the subject in the scene.

  • Name where the image will appear
  • Define the emotional job
  • Include brand boundaries
2 / What’s Current

V8.2 still the default, with published specs and a documented edit model

V8.2 took the default slot on July 24, 2026, the same day V8.0 alpha access ended. As of October 15, 2026 it’s still the default, and no V8.3 or V9 appears on any first-party page. But a long gap between versions means a quiet product, right? Not here. Five posts landed in September and three more on October 2 and October 8. They brought a lightbox editor, HD edit canvases, shared folders and a Thinking Mode test. And the documentation moves without announcements. The Version page carries no modification date, so a specification can appear there with no post at all. So watch the updates site and re-read the docs. The version number alone tells you very little.

2048 px
HD output documented for V8.2 by name, with SD at 1024 px
July 24, 2026
Last default-version change; no successor announced as of October 15, 2026
26 min
GPU time for one HD video batch of four, about $1.73 at the $4 per hour overage rate
$10 to $120
Monthly plan range, read October 15, 2026

V8.2 Default

Midjourney calls the V8.2 aesthetics “more creative, bold, sophisticated, edgy and fresh,” with far fewer unexpectedly low-quality outputs. Personalization draws on a much larger and improved pool of images. And no newer version exists on any first-party page. Thinking Mode is a way to rerun a V8.2 job, not a new version.
· Version documentation, read October 15, 2026, states plainly that “V8.2 released as the default version on July 24, 2026.”

  • Default version since July 24, 2026
  • Fewer weak frames in a batch
  • Improved personalization behavior

Published resolution specs Documented

HD is 2048 x 2048 px. SD is 1024 x 1024 px. Midjourney puts V8.2 next to those numbers, not 8.1, and it’s the only version-specific dimensions sentence on the page. The Version documentation backs it up, marking “HD Images (2048px)” against both V8.1 and V8.2. Here’s the catch, though. That page was rewritten to advance the version rather than to add one, and it carries no modification date, so a specification can land there with no announcement. So re-read the page. Don’t wait for a post to tell you it changed.
· Image Size and Resolution article, one sentence: “Midjourney version 8.2 creates HD images that are 2048 x 2048 pixels (px) and SD images that are 1024 x 1024px.”

  • HD 2048 x 2048 px, SD 1024 x 1024 px
  • Stated for V8.2 by name
  • The only remaining resolution gap is the edit model

V8.2 edit model Documented feature

It does three jobs. It edits an image from a plain-language instruction. It generates with up to four reference images “(replacing Omni Reference and Character Reference)”. And it drives inpainting and outpainting through the Editor. No first-party post announces a move out of testing, so treat it as documented rather than as a declared general release. Midjourney publishes neither pricing nor output dimensions for it.
· Edit Model article, filed under Using Your Own Images · described as compatible with Midjourney versions 8.1 and 8.2.

  • Instruction edits, references, inpaint and outpaint
  • Compatible with V8.1 and V8.2
  • No published pricing or resolution figures

Omni Reference V7 only

There’s no pending V8 omni-reference release to wait for. The V8 equivalent already shipped, and it’s the edit model. So any workflow parked on V7 for omni-reference reasons should be re-tested, not held.
· Omni Reference documentation, now explicit: “Omni Reference can only be used with Midjourney version 7.” V8.X users are pointed at “the Edit Model instead, our improved approach to editing and creating with reference images.”

  • V7 only, by documentation
  • Edit model is the named replacement
  • Re-test parked V7 recipes on V8.2

V8.1 Superseded

The default window and the announcement date look like they disagree. They don’t. The announcement post dates the switch itself. It shows June 11 because its timestamp is in UTC. In Pacific time it went up the evening of June 10, which is the date the docs give.
· Version documentation: V8.1 was the default from June 10 to July 23, 2026.
· Announcement post timestamp: 04:08 UTC on June 11, 2026.

  • Default June 10 to July 23, 2026
  • Announcement stamped June 11 in UTC
  • Re-test approved V8.1 routes against V8.2

HD is opt-in Toggle

HD is a setting, not a default. The August 21 changelog backs that up from the other direction, by fixing a case where Vary ran in HD without being asked. Here’s the newer wrinkle. Since September 24, 2026 a default set of parameters can be saved and will stick, so “on” and “off” are now somebody’s saved preference rather than a fresh choice each session. On alpha it goes one step further. Defaults can live per folder, and folders can now be shared. So open a shared folder and you may be running someone else’s settings.
· October 1 changelog, published October 2, 2026: “Prompt parameters are saved per folder, within a tab’s session.”
· Version documentation: “You can turn on HD images in the Version section of the settings panel on web, or by using the --sd and --hd parameters.”
· Same page, on cost: “HD costs 1.3 minutes of GPU time, compared to SD which costs 0.8 minutes of GPU time.”
· September 23 changelog, published September 24, 2026: you can now “save what’s in your prompt bar” as your defaults, from the Advanced settings panel or a “Set current pills as default” button.

  • Set it in the web settings panel
  • Or pass --hd per job
  • Saved defaults persist, so check them

Alpha changelogs September 3, 16 and 24, 2026

Alpha is no longer a rough draft with one fix in it. The September 16 changelog added Korean language support, editing and prompt-bar improvements, mobile and tablet optimization, hover preview for source images, and an Enhance option replacing Upscale. The September 23 changelog, published September 24, takes Korean to every web user, adds savable default parameters, and lets edit canvases run in HD. No new model version in any of them. One catch on the HD canvas. The Version documentation still says HD inpainting and outpainting come back at SD size. So measure what you get.
· September 23 changelog: “Now, it’s live for everyone on midjourney.com.” And: “The editor now supports v8.1 and v8.2 edit types, and edit canvases can run in HD.”
· September 3 changelog: “the new v8.2 edit model is on alpha, with a new editor built directly into the lightbox”. In it you “open any image in the lightbox, edit with plain instructions, attach up to 4 reference images, and see all your session edits in one place”.

  • Lightbox editor with up to four references
  • Korean now live for all web users
  • HD canvas in the editor for V8.1 and V8.2

Edits stopped degrading the image September 24, 2026

This is the fix that changes how you work, not just what you click. Repeated local edits used to cost you quality. They don’t now.
· “Making targeted edits to the image will change only the pixels that you’ve selected. This gives you more control, and allows you to make repeated edits without degrading the quality of the image.”
· Same post, on the tile parameter: “We’ve also made tile --tile work better for V8.1/8.2. Previously, you might have seen seams between your tiled images. Now, they blend invisibly from one tile to the next.”
· Also shipping: live style previews on alpha, so the Styles sidebar can show your current prompt across styles before you spend on it.

  • Repeated inpaint and outpaint keeps quality
  • --tile seams fixed on V8.1 and V8.2
  • Live style previews for cheap exploration

Thinking Mode Testing, October 8, 2026

This is new, and it isn’t a version. You take a finished V8.2 job, standard or edit, and rerun it with “thinking.” Midjourney says the model checks where the image missed the prompt and generates again. It’s an early test on alpha. And here’s what nobody has published: the GPU cost, which plans get it, or a release date. So budget a thinking rerun as at least one more generation until Midjourney says otherwise. That’s an AI Mindset reading, not a vendor figure.
· Testing Thinking Mode, October 8, 2026: click “Rerun (Thinking)” under the lightbox for a job. “We’re finding in tests that this helps make the model better at prompt accuracy, typography, and coherence,” a vendor claim. On timing: “In the future we might make this available broadly.”

  • A rerun for V8.2 standard and edit jobs
  • No published cost, plan gating or date
  • Budget it as another generation

Shared folders Alpha, October 8, 2026

The collaboration Midjourney promised in September shipped on alpha. A folder can be shared by username, with Collaborators who generate into it and Viewers who browse. It can also go out by link, stay private, or open to any Midjourney member. The owner can delete a shared folder for everyone, and images stay in each creator’s history. This is the sanctioned way to work as a team, so there’s no excuse for a shared login. But read the privacy line twice. Sharing a folder doesn’t hide anything. Non-stealth images can still show up on Explore and on profiles.
· October 7 changelog, published October 8, 2026: “folders can now be shared!” and “sharing doesn’t change image privacy.”

  • Collaborator and Viewer roles
  • Sharing doesn’t make work private
  • Use it instead of a shared account

Two surfaces, both live Web and Discord

Midjourney runs on the website and in Discord, and neither one is marked for retirement. That matters for two reasons. Custom suffixes are marked unavailable on the web. Option sets can be used in web prompts, but they can only be created and managed in Discord. And a team working in a shared Discord server is exposed by default, because the terms say anything you make in a shared or open space such as a Discord chatroom is viewable by anyone in that chatroom.
· Web vs Discord article: “With your Midjourney subscription, you have the freedom to create images on both our website and Discord.” Same page: “Most features and tools are offered on both platforms, but sometimes they might have different names.”

  • Neither surface is deprecated
  • Custom suffixes are Discord only
  • Shared Discord rooms expose work to the room

Roadmap input: the voting session is overdue September 3, 2026

Midjourney is asking for feature ideas directly, at midjourney.com/ideas. Then it promised a formal voting session inside a week or two. The outer bound of that window was September 17. As of October 15, 2026 it’s 42 days since the promise and 28 days past that outer bound, and nothing has been published either way: no vote, no result, no reschedule, no withdrawal. None of the three posts on October 2 and October 8 mentions a vote. Their “Up next” lists name collaborative sharing, edit history and style tools instead. Treat it as pending and overdue rather than dropped. Got a blocking gap? The ideas page is still a cheap way to put it in front of the vendor. Just don’t plan around a voting date.
· September 3 post: “submit your ideas here and we will do a formal voting session in the next week or two.”

  • Ideas collected at midjourney.com/ideas
  • Outer bound of the promise expired September 17
  • No first-party page records a vote either way

First acquisition July 23, 2026

Midjourney bought Co-Star. Midjourney says Co-Star stays entirely under founder Banu’s control and direction, and will help build a broader organization around Product and Design.

  • David Holz frames Midjourney as “a very unusual research lab”
  • Half a dozen similarly ambitious projects teased in July for the following six months
  • Midjourney Medical is already underway

V8.0 Alpha Withdrawn July 24, 2026

That’s a hard end date, and it lands on the same day V8.2 became the default. V8.0 isn’t available anywhere now.
· Version documentation: V8.0 “was an early access alpha version that launched on March 17, 2026 and was available on the Midjourney Alpha website until July 24, 2026.”

  • Available March 17 to July 24, 2026 only
  • Do not plan any workflow around alpha access
  • Re-test archived alpha recipes in V8.2

Video generation

Midjourney turns a still into a five-second clip and returns four options by default. You get low and high motion, loop, end-frame and batch controls. Clips extend in four-second steps, up to four extensions, for a maximum length of 21 seconds. So a 21-second clip is five generations of spend, not one.
· Video documentation, the budgeting line that matters: “Extending a video costs the same amount of GPU time as an initial video generation.”

  • Five seconds initially, 21 seconds maximum
  • Up to four extensions of four seconds each
  • Each extension costs a full generation

Four core tiers

Basic, Standard, Pro and Mega are the core subscription tiers. What separates them is Fast time, Relax access, privacy and concurrency. The table row reads “General Commercial Terms” on all four tiers. A footnote under it carries the revenue test, but as a buying rule. It doesn’t say what happens to ownership if you ignore it. Only the terms say that.
· Comparing Midjourney Plans article, read October 15, 2026, footnote: “You must purchase the Pro or Mega plan if you are a company making more than $1,000,000 USD in gross revenue per year.”

  • Standard adds unlimited Relax images
  • Pro and Mega add SD Relax video and Stealth
  • Extra GPU time is $4 an hour on every tier

Which version should you use?

Start with V8.2 unless a specific compatibility need says otherwise.

New image work
Use V8.2

It’s the default general image model in the product as of October 15, 2026. You get the bolder aesthetic and the improved personalization pool. Published dimensions are 2048 x 2048 px in HD and 1024 x 1024 px in SD. Nothing newer exists.

  • Start in SD
  • Use Raw when needed
  • Turn HD on for winners
3 / Image System

Consistency comes from references and curation, not longer prompts

When output drifts, the instinct is a longer prompt, right? That’s not the answer. V8.2 is bolder by default and pulls personalization from a larger image pool, but control still comes from a deliberate reference system, documented settings and a human choosing. The edit model adds instruction-based editing and a four-image reference slot to the control surface. That doesn’t change the principle.

Instruction

Natural-language brief

Describe subject, scene, visual language and use. The V8 line is designed to read detailed prompts more coherently than earlier versions.

Adherence

Raw mode

Turn down Midjourney’s default styling when the brief or the reference should lead. Useful when the V8.2 default look is stronger than the brand allows.

Taste

Personalization

Rate images to pull output toward one person’s or one team’s visual preference. Personalization is documented as compatible with versions 6 and later, and V8.2 draws on a much larger pool of images. Mind the one-way rule. A Global V7 profile works in V8.1 and V8.2, but V8 profiles aren’t compatible with V7.

Style

Style references

Carry a visual language across a set without trying to describe every aesthetic property in words.

Direction

Moodboards

Build a curated visual world that informs several prompts and campaign routes.

Variation

Remix and variants

Change the prompt or the settings while keeping useful structure from the selected generation. As of August 21, 2026, Vary respects your settings instead of quietly running in HD.

Local edit

Vary region, inpaint, outpaint

Fix one area instead of throwing away a strong composition. On V8.X, inpainting and outpainting now use the edit model, so a masked fix and an instruction edit run through the same system. Since September 24, 2026 a targeted edit changes only the pixels you selected, which means you can edit the same frame repeatedly without grinding the quality down. The Version documentation still says HD inpainting and outpainting downscale to SD, though. So measure the delivered size.
· Editor documentation, plus the September 24, 2026 edit updates post.

Resolution

HD and SD switch

Set HD in the Version section of the web settings panel, or pass --hd or --sd on the job. Neither one is automatic. V8.2 HD is documented at 2048 x 2048 px and SD at 1024 x 1024 px.

Diagnose the weak result

The next move depends on what is actually wrong.

The idea is weak
Return to exploration

Do not edit a weak idea into submission. Change the metaphor, the scene or the communication strategy, and generate new directions.

  • Stay in SD
  • Change conceptual variables
  • Review against audience intent
4 / Video

Lock the still before you pay to animate it

Motion is the expensive part, right? So the still has to be right first. Midjourney video works as a second-stage motion tool: it extends an approved image into a short clip. It’s not a substitute for storyboarding or post-production. Midjourney publishes exact output dimensions, exact per-batch GPU minutes and a flat $4 an hour for extra GPU time. So you can put a dollar figure on a clip before you generate it.

The cost-aware video workflow

Use motion only after the frame is doing its job.

Lock frame
Approve the image first

Resolve subject, styling, framing, text space and brand questions in the still workflow.

  • Do not animate unresolved concepts
  • Use an approved seed
  • Check product truth
Aspect ratioSD outputHD output
1:1624 x 624 px960 x 960 px
16:9832 x 464 px1280 x 720 px
4:3720 x 544 px1104 x 832 px
2:3512 x 768 px784 x 1168 px
1:2448 x 880 px672 x 1360 px
Batch sizeSD GPU minutesHD GPU minutesHD cost at $4 per hour of overage
Four videos8 minutes26 minutesabout $1.73
Two videos4 minutes13 minutesabout $0.87
One video2 minutes7 minutesabout $0.47

Low motion

Use it for ambient movement, gentle camera shifts and image-preserving animation.

  • Editorial loops
  • Subtle social assets
  • Lower risk of warping

High motion

Use it when dynamic subject or camera movement is central to the idea and some instability is acceptable.

  • Action and transformation
  • More physics stress
  • Review subject consistency

Loop and end frame

Use explicit loop or ending control when the channel or the sequence needs continuity.

  • Plan the transition
  • Check the first and last frame
  • Do not rely on accidental loops

SD video

SD video is 624 x 624 px at 1:1 and 832 x 464 px at 16:9. On Basic and Standard, every video frame comes out of Fast time.
· Video documentation, precise about who can run it: “All plan tiers can generate videos in Fast Mode, but only Pro and Mega plans can generate videos in Relax Mode.”

  • 8 GPU minutes for a batch of four
  • Fast Mode on every tier
  • Relax video is Pro and Mega only

HD video

HD video is 960 x 960 px at 1:1 and 1280 x 720 px at 16:9. It runs on Standard, Pro and Mega, in Fast mode only. One batch of four costs 26 GPU minutes. On Standard, the lowest tier that can run it, that’s just under 3 percent of the 15 monthly Fast hours. Which is why HD video belongs after the motion direction is approved.

  • 26 GPU minutes for a batch of four
  • 7 minutes for a single video
  • Use after motion direction is approved

Batch size

Current video generation returns four outputs by default and supports video-specific batch controls. Dropping to a batch of two cuts SD from 8 GPU minutes to 4 and HD from 26 to 13. Dropping to one cuts them to 2 and 7. That’s the cheapest dial on the page, and most teams never touch it.

  • Compare motion outcomes
  • Do not select from the thumbnail alone
  • Track the cost per approved clip
5 / Prompt Lab

Write a creative brief, not a keyword inventory

Stop collecting keywords. The V8 line rewards coherent natural language. The best prompt says what the asset must communicate and how the frame should behave. Piling up aesthetic tokens does none of that.

Build the prompt around the job

Choose the asset to see the information Midjourney needs most.

Campaign hero
Prioritize idea, hierarchy and copy space

State the campaign message, the audience, the hero subject, the emotional tension, the composition and where copy must live.

  • One visual idea
  • Clear focal hierarchy
  • Channel-specific aspect ratio
Campaign hero
Campaign hero for a report about overconfidence in AI: one executive at the edge of a brightly lit stage, the audience receding into darkness, restrained corporate realism, generous negative space at upper left for headline, 16:9.
Product visualization
Premium matte-black travel speaker, exact cylindrical proportions from the reference image, on a pale stone table in a modern hotel room at dawn, soft side light revealing material texture, three-quarter hero angle, no extra controls or logos.
Editorial metaphor
Editorial illustration about hidden automation risk: a pristine office tower whose foundation is a dense, tangled network of cables, authoritative geometric composition, screen-print texture, navy and controlled orange palette, vertical magazine opener.
Edit instruction
Using the approved frame, change only the lighting to late-afternoon side light and extend the canvas to the left for headline space. Keep the subject, wardrobe, product geometry and color palette identical.
Motion brief
Animate this approved still: subject remains nearly motionless; papers lift gently in the ventilation; camera makes a slow three-percent push-in; low motion; no face change; seamless five-second loop.
6 / Plans

Choose by throughput, privacy and video, not aspiration

Pick the plan on behavior, not ambition. How much Fast time does the team burn? Can it wait for Relax? Does the work have to be private? Is video part of the normal week? And is the company over $1,000,000 in revenue, because above that line the plan decides whether you own the output at all.

PlanMonthlyAnnualFast GPURelaxPrivacy
Basic$10$96, i.e. $8/month3.3 hr/month (200 minutes)NonePublic workflow
Standard$30$288, i.e. $24/month15 hr/monthUnlimited imagesPublic workflow
Pro$60$576, i.e. $48/month30 hr/monthUnlimited images + SD videoStealth Mode
Mega$120$1,152, i.e. $96/month60 hr/monthUnlimited images + SD videoStealth Mode

Find the operational plan

Pick the behavior that best matches the team.

Regular image production
Standard

Use Standard when unlimited Relax images matter and video privacy isn’t yet a routine requirement. Standard is the lowest tier with HD video, but all of its video runs on Fast time.

  • 15 Fast hours
  • Unlimited Relax images
  • No Stealth Mode

You own the output, with exceptions

You own what you make, right? Mostly. And ownership doesn’t lapse when the subscription does. But read the next sentence of the contract, not just the first one. The ownership clause doesn’t end where the quote usually ends.
· Terms of Service, effective May 27, 2026: “You own all Assets You create with the Services to the fullest extent possible under applicable law.” The clause goes straight on to list exceptions, including third-party rights and upscaling someone else’s image.
· The commercial-use article adds the part buyers actually ask about: “With Midjourney, you own all the images and videos you create, even if you decide to cancel your subscription.”
· Prices: Comparing Midjourney Plans, Midjourney docs, October 15, 2026.

  • Ownership stated in the terms
  • Survives cancellation
  • The terms name exceptions, including a revenue one

The license you grant back

Ownership isn’t exclusivity. You grant a license back to Midjourney, and it covers your inputs as well as your outputs. Working for a regulated client, or under a confidentiality clause? That’s the sentence to put in front of legal before the first upload.
· The same terms have you grant Midjourney “a perpetual, worldwide, non-exclusive, sublicensable no-charge, royalty-free, irrevocable copyright license.”

  • Perpetual and irrevocable
  • Covers prompts and reference uploads
  • Sublicensable by Midjourney

Public work can be remixed

On any plan without Stealth Mode, remixable is the default condition of the work. Not an edge case. Stealth is available only on Pro and Mega.
· Midjourney describes itself in the terms as “an open community which allows others to use and remix Your Content whenever they are posted in a public setting.”

  • Public posts are remixable by others
  • Stealth Mode on Pro and Mega only
  • Decide privacy before the first client brief

Upscaling isn’t authorship

Anything pulled from the public feed and finished in-house is somebody else’s asset. However much work went into the finish.
· The commercial-use article is direct: “If you upscale an image that another user created, that image belongs to the original creator, not you.”

  • Upscaling grants no ownership
  • Check provenance before production
  • Keep a source record per asset

The $1M condition is an ownership rule

This is the sentence to put in front of legal first. It isn’t a commercial-use permission. It’s a condition on ownership itself, and the plan you are on decides it. A $1,000,000 company sitting on Basic or Standard doesn’t own what its people generate. The threshold is company revenue, not Midjourney spend, so a two-person team inside a large business is already over it.
· Terms of Service, effective May 27, 2026: “If you are a company or any employee of a company with more than $1,000,000 USD a year in revenue, you must be subscribed to a ‘Pro’ or ‘Mega’ plan to own Your Assets.”
· Worth knowing before counsel finds it: three first-party pages word the test three ways. The terms say “more than $1,000,000 USD a year in revenue,” with no qualifier, and they cover employees too. The commercial-use article says “a business grossing more than $1,000,000 USD a year.” The plan comparison footnote says “a company making more than $1,000,000 USD in gross revenue per year.” Only the terms tie it to ownership. Assume the broader reading until Midjourney says otherwise.

  • Ownership, not just commercial permission
  • Pro or Mega required above the line
  • Three first-party pages, three wordings

One account per person

A shared studio login is a contract breach, not a cost saving. The terms are flat about it, and teams assume the opposite on a Pro or Mega seat because the allowances look big enough to share.
· Terms of Service, effective May 27, 2026: “Only one user may use the Services per registered account. Each user of the Services may only have one account.”
· Enforcement is visible, too. The October 7 changelog, published October 8, 2026, says “Temporary suspensions come with an Appeal button,” and restricted accounts see dimmed controls with an explanation. For team work, alpha now has shared folders.

  • One person per account, both directions
  • Budget a seat per creator
  • Shared folders, not shared logins

Extra GPU time is $4 an hour

Same price on every tier. That’s what turns a GPU minute into a number you can put in a quote. An HD video batch of four is 26 minutes, so about $1.73. A full 21-second HD clip is five generations, so about $8.67. An HD image costs 1.3 minutes against 0.8 for SD, which is under a dime either way. But bought hours don’t keep. They expire after 60 days, and they only work while a subscription is active. Monthly plan hours don’t roll over at all. So buy what the job needs, not a stockpile. Put the date on the calendar, re-read the source page on the same cadence, and keep a dated copy of the row you depend on. A date written down once and trusted is how you miss a deadline that moved.
· Comparing Midjourney Plans article, “Purchase Extra GPU Time” row, read October 15, 2026.
· Purchasing Extra Fast Time: “these purchased hours come with a 60-day expiration.” Subscription Fast Time Expiration: “any unused Fast time from your subscription will not carry over to the next month.”

  • $4 per hour on Basic, Standard, Pro and Mega
  • Bought hours expire after 60 days
  • Price the clip before you brief it

Relax is a queue, not a promise

Unlimited Relax sounds like free capacity. It’s a waiting line. Midjourney says Relax waits usually run 0 to 30 minutes and depend on how much Relax you’ve used next to everyone else. And some work can’t run there at all: permutations, --repeat, HD video and Max Upscale. So a deadline job belongs on Fast time. Turbo is the other way round. It burns double the Fast time per image, it’s experimental, and it isn’t supported in V8.1.
· GPU Speed (Fast, Relax, Turbo) article, read October 15, 2026: Turbo “uses double the Fast time for each image.”

  • Relax waits depend on your own usage
  • No HD video, permutations or repeats in Relax
  • Turbo costs double Fast time

There is no API, and no automating around it

No API. No waitlist. No partner program. As of October 15, 2026, fifteen months after the one exploratory survey, there’s no follow-up post. And here’s the part that closes the obvious workaround: the terms prohibit automating the web app at all. So plan integration around people using the product, not around a script that drives it.
· Terms of Service, effective May 27, 2026: “You may not use automated tools to access, interact with, or generate Assets through the Services.” The same restrictions list bars reverse engineering the Services or the Assets, reselling or redistributing access, and using the Services to develop or offer competing products.
· Enterprise API Survey, July 16, 2025, the only first-party artifact: Midjourney is “starting to investigate opening up an Enterprise API,” and it points at an application form.
· Midjourney’s docs describe no API, Teams or Enterprise plan, and the homepage mentions neither.

  • No endpoint, no waitlist, no partners
  • Automated access is contractually prohibited
  • Confirm before promising any automation
7 / Behavior Playbook

Eight habits that turn Midjourney into a production system

You’d expect the model to separate good work from bad, right? It doesn’t. The team does. Better briefs, consistent references, documented selection, deliberate spend. That’s the advantage, and it holds.

01

Name the visual job

Define the audience, the channel and the communication goal before you describe the scene.

02

Explore real breadth

Generate conceptually different routes rather than twenty cosmetic variants.

03

Curate against strategy

Choose the idea that works, not the image that looks expensive.

04

Build a reference system

Use style references, moodboards and personalization consistently across a campaign. Check version compatibility before you standardize on a profile.

05

Change one variable

Make iteration legible by changing composition, style or subject on purpose.

06

Edit locally

Protect a strong composition by fixing the smallest wrong region. Midjourney publishes no resolution or pricing figures for the edit model, so benchmark it against your own work before it carries a client deadline.

07

Spend after approval

Switch HD on and generate video only once the route is selected. One HD video batch is 26 GPU minutes, about $1.73 at the $4 an hour overage rate.

08

Archive the recipe

Store prompt, version, seed, references, parameters and final asset together. Record whether HD was on, whether a saved or folder default put it there, and whether the edit model or a thinking rerun touched the frame.

8 / Watch Outs

Beautiful output can still be strategically, factually or legally wrong

A beautiful image hides a wrong one, right? Visual persuasion is efficient that way. So review has to go past whether the image looks good, and you need to know which published figures exist and which don’t.

The edit model has no numbers

Dimensions are published for V8.2. The edit model is the exception. Neither the Edit Model article nor the Editor article carries an output dimension, a GPU-minute figure or a price. So measure your own edit output before you write it into a delivery spec. The sources pull in two directions here. The September 23 changelog says edit canvases can run in HD. The Version documentation still says HD inpainting and outpainting come back at SD, and it doesn’t say whether that note covers V8.2. No first-party post declares the feature out of testing either, so it’s documented rather than formally released.
· Version documentation, read October 15, 2026: “Using any of our inpainting or outpainting tools (Pan, Zoom Out, Edit/Vary Region) on HD images will downscale the resulting images to SD.”

A V7-only reference path

Omni Reference is documented as working only with version 7, and Midjourney names the edit model as its V8.X replacement. So a recipe held on V7 to keep omni-reference behavior is parked on a superseded path, not waiting for a promised upgrade. Re-test those reference sets on V8.2.

Personalization runs one way

A Global V7 profile works in V8.1 and V8.2, but V8 profiles aren’t compatible with V7. Build a V8 profile, then need to reproduce a V7 asset, and you can’t carry the profile backward. So record which version a profile was built in, alongside the prompt and the seed.

Editor gating is unstated

The Editor documentation doesn’t say which plan tiers may use it. The plan comparison table covers Fast time, Relax, privacy and concurrency, and it isn’t a safe proxy for editor access. Confirm with Midjourney before you promise it on a Basic or Standard seat.

Two dates, one moment

The V8.1 default looks like it has two dates. It doesn’t. The announcement post shows June 11 because its timestamp is in UTC, and in Pacific time it went up the evening of June 10. The docs say June 10. So write June 10, and note the UTC stamp if anyone cites the post in a contract.
· V8.1 announcement post, stamped 04:08 UTC on June 11, 2026. The Version documentation says V8.1 was the default from June 10.

HD assumptions

Don’t assume HD is on. It’s a toggle in the web settings panel, or the --hd and --sd parameters, and a budget model built on an assumed default will be wrong. The August 21 changelog fixed a case where Vary ran in HD without being asked, which is exactly the kind of silent cost the assumption produces. And since September 24, 2026 a default set of parameters can be saved and stays saved. On alpha, defaults can also live per folder, and folders can be shared. So someone else’s preference can now be running on your job.

Ownership has a revenue condition

The headline is that you own your output. The contract qualifies it. A company with more than $1,000,000 USD a year in revenue, or an employee of one, must be on Pro or Mega to own its assets, and the ownership clause itself names exceptions. The plan comparison page does show the threshold, in a footnote. But it words it as a buying rule and says nothing about ownership. So check the revenue test against the terms, not the help center or the plan page.

Default drift

The default model changed twice in roughly six weeks: V8.1 on June 10, V8.2 on July 24. As of October 15, 2026 there’s still no V8.3 or V9 on any first-party page, so the gap between versions has stretched while the product keeps moving in other ways. Prompts approved under V8.1 may still render differently under V8.2. So pin and re-test the workflows that matter.

Public creation

Most plans don’t provide Stealth Mode, and Midjourney describes itself in its terms as an open community where public content can be used and remixed by others. Discord compounds it, because the terms warn that anything made in a shared or open space such as a Discord chatroom is viewable by anyone in that chatroom. That exposure is independent of your plan. Shared folders don’t change it either. Midjourney says plainly that “sharing doesn’t change image privacy,” so a private-looking team folder can still feed Explore and profiles. Confirm privacy, and the room, before you use client references or unreleased products.

Product hallucination

Generated products can change controls, labels, proportions or materials while still looking photorealistic.

Text accuracy

The V8 line improves rendering, but generated packaging, UI and copy still need inspection and often manual replacement.

Likeness and rights

You own your outputs under the terms effective May 27, 2026, subject to the revenue condition above. You also grant Midjourney a perpetual, sublicensable, irrevocable license over your inputs and outputs. And ownership doesn’t remove trademark, likeness, confidentiality or sector-specific risk. Instruction-based editing of an existing image raises the question of whose image it was in the first place, and upscaling another user’s image doesn’t make it yours.

Brand drift

A strong aesthetic isn’t automatically the organization’s aesthetic, and V8.2 is described as bolder and edgier by default. References and curation have to carry the brand.

GPU waste

High-resolution generation and video eat capacity fast when the creative decision is still open. One HD video batch is 26 GPU minutes, and every four-second extension costs as much as the original generation. A thinking rerun has no published cost at all, so treat it as another generation until Midjourney prices it.

Stakeholder bias

Polished images close discussion too early. Present routes with the strategic rationale, not as finished answers.

Company focus

Midjourney describes itself as a research lab and teased half a dozen ambitious projects in July for the following six months, with Midjourney Medical already underway. That window runs to about January 23, 2027. Its homepage now lists Image and Video Models, Medical, and six unnamed projects, three in software and three in hardware. So roadmap attention may not sit entirely with image generation.
· midjourney.com homepage, a rolling page read October 15, 2026: “a community-funded research lab of 60 people,” with six projects marked TBA.

9 / Sources

Where these facts come from

Every fact comes from a Midjourney page read on October 15, 2026. Here’s the catch. No article on docs.midjourney.com prints a modification date, so each documentation citation is a dated reading, not a dated page. The Terms of Service, effective May 27, 2026, is the one exception. So use one rule. A dated article is authoritative for what was announced and when. A live page is authoritative for what is true today. When they disagree, the live page governs the present and the article governs the record. Keep your own dated copy of the live page, because it carries no history. The updates site carried five posts in September and three in October, and all eight are cited below.

Version 8.2Midjourney · July 24, 2026 · checked October 15, 2026 · the newest version post on updates.midjourney.comImage Size and ResolutionMidjourney · undated documentation, checked October 15, 2026 · “Midjourney version 8.2 creates HD images that are 2048 x 2048 pixels (px) and SD images that are 1024 x 1024px”Version documentationMidjourney · undated rolling reference, checked October 15, 2026 · version dates for V8.0, V8.1 and V8.2 · HD opt-in at 2048px, with HD inpainting and outpainting downscaled to SDEdit ModelMidjourney · undated documentation, checked October 15, 2026 · instruction edits and up to four references “(replacing Omni Reference and Character Reference)” · versions 8.1 and 8.2 · no pricing or output dimensions · slug still reads Character-ReferenceOmni ReferenceMidjourney · undated documentation, checked October 15, 2026 · “Omni Reference can only be used with Midjourney version 7,” with V8.X users directed to the Edit ModelEditorMidjourney · undated documentation, checked October 15, 2026 · inpainting and outpainting use the edit model on V8.X · no plan gating statedEdit Model for V8Midjourney · August 27, 2026 · checked October 15, 2026 · the dated origin of the V8.2 image edit model, announced for open testingTesting Thinking ModeMidjourney · October 8, 2026 · “Rerun (Thinking)” for V8.2 jobs on alpha, claimed to help prompt accuracy, typography and coherence · no cost, plan gating or release dateAlpha Changelog, October 7, 2026Midjourney · October 8, 2026 · shared folders with Collaborator and Viewer roles · “sharing doesn’t change image privacy” · thinking reruns for 8.2 jobs · an Appeal button on temporary suspensionsAlpha Changelog, October 1, 2026Midjourney · October 2, 2026 · prompt parameters saved per folder · multiple style references as one pill · edits keep the base image aspect ratioAlpha Changelog, September 23, 2026Midjourney · September 24, 2026 · checked October 15, 2026 · Korean “live for everyone on midjourney.com” · savable default parameters · v8.1 and v8.2 edit types in the editor, with HD canvases · no voting sessionEdit updates, thumbnail previews, and moreMidjourney · September 24, 2026 · checked October 15, 2026 · targeted edits change “only the pixels that you’ve selected” · --tile seams fixed for V8.1 and V8.2 · live style previews on alphaAlpha Changelog, September 16, 2026Midjourney · September 16, 2026 · checked October 15, 2026 · Korean on alpha, editing and prompt-bar improvements, mobile and tablet work, hover previews, Enhance replacing UpscaleAlpha Changelog, September 2, 2026Midjourney · September 3, 2026 · checked October 15, 2026 · the v8.2 edit model on alpha in a lightbox editor, with up to four reference images per editCollecting ideas to improve MidjourneyMidjourney · September 3, 2026 · checked October 15, 2026, unmodified · ideas at midjourney.com/ideas, with “a formal voting session in the next week or two” promisedEnterprise API SurveyMidjourney · July 16, 2025 · checked October 15, 2026 · Midjourney is “starting to investigate opening up an Enterprise API” · no follow-up post as of October 15, 2026V8.1 is now the defaultMidjourney · stamped 04:08 UTC on June 11, 2026, the evening of June 10 in Pacific time · checked October 15, 2026V8.1 updatesMidjourney · 2026 update log · origin of the “2K images, without upscaling” claim for V8.1, superseded as a specification source by the Image Size and Resolution articleMidjourney’s first acquisitionMidjourney · July 23, 2026 · checked October 15, 2026 · Co-Star, plus half a dozen ambitious projects teased for the following six monthsComparing Midjourney PlansMidjourney · undated documentation, checked October 15, 2026 · prices, allowances and $4 extra GPU time · footnote: Pro or Mega above “$1,000,000 USD in gross revenue per year”GPU Speed (Fast, Relax, Turbo)Midjourney · undated documentation, checked October 15, 2026 · Relax eligibility and typical waits · features unavailable in Relax · Turbo at double Fast time and unsupported in V8.1Purchasing Extra Fast TimeMidjourney · undated documentation, checked October 15, 2026 · purchased hours expire after 60 days and need an active subscriptionSubscription Fast Time ExpirationMidjourney · undated documentation, checked October 15, 2026 · unused subscription Fast time doesn’t carry over to the next monthVideo documentationMidjourney · undated documentation, checked October 15, 2026 · clip length, extension cost, SD and HD dimensions and GPU minutes per batchPersonalizationMidjourney · undated documentation, checked October 15, 2026 · versions 6 and later · “Your Global V7 Profile works in V8.1 and V8.2. V8 profiles are currently not compatible with V7”Terms of ServiceMidjourney · effective May 27, 2026 · checked October 15, 2026 · ownership, its $1,000,000 condition, the license granted back, the automated-access ban and one account per personUsing Images and Videos CommerciallyMidjourney · undated documentation, checked October 15, 2026 · ownership survives cancellation · upscaling another user’s image transfers nothing · the $1,000,000 test worded as “a business grossing more than $1,000,000 USD a year”Getting Started GuideMidjourney · undated documentation, checked October 15, 2026 · names no API, enterprise offer or automation routeWeb vs DiscordMidjourney · undated documentation, checked October 15, 2026 · both surfaces live · custom suffixes marked unavailable on the web · option sets usable in web prompts but created and managed only in Discord

AI Mindset

Explore the model cheatsheets