Midjourney V8.2 Features and the Flags V8 Retired

Midjourney V8.2 shipped with a launch note that runs to four bullet points. The feature compatibility chart sitting in the docs runs thirty-four rows, and nine of them are marked unsupported for this generation^[1]. Those nine rows are where the upgrade actually bites, because they include flags that most Midjourney tutorials written before 2026 still teach as standard.

Midjourney V8.2 became the default render on July 24, 2026^[1]. If you set a version and forgot about it, you are already on it. This piece covers what the release changed, which parameters stopped working somewhere between V6 and V8, how the Personalization system behaves across versions, and the jobs where the V8.2 look is the reason to reach for it.

TL;DR

• Midjourney V8.2 became the default version on July 24, 2026, and the release focuses on aesthetics, image quality and Personalization^[1].
• Midjourney says the odds of a random low-quality render "should be dramatically reduced," which matters more for production work than any single new capability^[2].
• Nine features are unsupported on V8.1 and V8.2: Character Reference, Character Weight, Omni Reference, Omni Reference Weight, Multi-Prompting, Quality, Niji, Draft Mode and Turbo Mode^[1].
• There is no Global V8 Personalization profile. Your V7 Global profile works in V8.1 and V8.2, but profiles you build for V8 will not run in V7^[3].
• Style Reference, Style Weight, Stylize, Chaos, Raw, Seed, Weird and --exp all survive into V8, so steering happens through those rather than through character flags^[1].
• HD renders at 2048px and costs 1.3 minutes of GPU time against 0.8 for SD, and any inpaint or outpaint on an HD image downscales it back to SD^[1].

What Midjourney V8.2 actually shipped

The announcement is short enough to quote almost in full. Midjourney describes V8.2 as an update focused on aesthetics, image quality and personalization, then lists four effects: images should be more creative, bold, sophisticated, edgy and fresh; random low-quality results should be dramatically reduced; Personalization will understand personal taste much better, especially on profiles with many ratings; and building a Personalization profile for V8.2 now draws from a much larger, improved pool of images^[2].

Read the second bullet again. Aesthetic upgrades are the ones that get demoed, but a reduction in random duds is the one that changes how you budget a job. If one render in six comes back unusable, a four-image grid gives you roughly three viable takes instead of four, and the cost of a finished asset moves accordingly. Midjourney published no figure for the improvement, so treat it as a direction rather than a number.

The version doc adds the surrounding timeline. V8.1 shipped April 14, 2026 and held the default slot from June 10 to July 23, 2026, arriving as the fastest model in the lineup with standard jobs rendering roughly four to five times faster than earlier versions^[1]. V8.2 replaced it the next day. Both remain selectable, which makes the choice between them a real one rather than a legacy fallback.

The parameters V8 quietly retired

This is the part the release note does not cover. Midjourney maintains a feature compatibility chart comparing V6, V7 and the V8.1/V8.2 pair, and it is the least ambiguous document the company publishes on the subject^[1].

Four of these deserve a note.

Quality is gone. --q accepted 0.5, 1 and 2 on V6, then 1, 2 and 4 on V7. On V8 it is unsupported outright. Any prompt template carrying --q 2 is passing a flag the model no longer reads.

Turbo is gone. V6 and V7 both offered Turbo Mode. V8.1 and V8.2 do not. Fast and Relax survive, so the speed decision on V8 is a two-way choice, not three.

Multi-prompting is gone. The :: weighting syntax that let you split a prompt into weighted components was already dropped in V7 and stays unsupported in V8.

Omni Reference runs a different model. The chart marks it "Uses V7" for V8.1 and V8.2. Attaching an omni image does not give you a V8 render with a reference on it. It gives you a V7 render. If output from a referenced prompt looks unlike the V8 work you produced an hour earlier, the version changed underneath you.

What survived is substantial: Variations, Remix, Moodboards, Style Reference and its codes, Style Weight, Image Prompts, Image Weight, the --no parameter, Seed, Chaos, Raw, Stylize, Tile, Weird and --exp^[1]. Pan, Zoom Out and the Editor still work but run through V6.1 rather than V8. Seed carries an asterisk: the chart describes V8 seeds as 99% identical rather than exact, so seed-locked reruns land close but not on top of each other.

Personalization is the headline, and it has a one-way version trap

Midjourney V8.2 puts most of its weight behind Personalization, and the mechanics are worth understanding before you build anything on top of it.

Personalization works as a style profile trained on images you select. You unlock a Global profile by picking favourites from a grid until a progress bar fills, then apply it with --p, or with --p pID for a specific profile^[3]. Midjourney converts the profile ID to a code at submit time, which is why old prompts show codes rather than IDs.

Here is the constraint that catches people. There is currently no Global V8 profile. Your V7 Global profile is what runs when you use Personalization on V8.1 or V8.2. You can create additional profiles for V8.x, but those V8 profiles are not compatible with V7^[3]. Compatibility runs one way only.

The practical consequences:

• A team standardising on Midjourney V8.2 can build V8 profiles freely, and should accept that those profiles pin them to V8.
• Anyone still switching between V7 and V8 for different jobs needs the V7 Global profile as the shared baseline, since it is the only one both versions read.
• Liking images on the Explore page feeds your Global profile for whichever version produced that image, so browsing habits quietly shape your defaults^[3].

Midjourney's own advice is to spend longer selecting images during unlock, and the V8.2 note repeats it: profiles with a large number of ratings benefit most from the new model^[2]. A profile unlocked in five minutes and never touched again is not the thing the release was tuned for.

Style reference and stylize do the steering

With character flags off the table, consistency on Midjourney V8.2 comes from stacking the controls that remain.

Style Reference captures the visual character of an image and applies it to new work. It carries colour, medium, texture and lighting rather than objects or people, and it is compatible with V6 and later^[4]. Midjourney's documented best practice is counterintuitive if you come from instruction-style prompting: describe the content you want, not the transformation. "Detailed portrait of a dog" is the recommended form. "The look of this image but a dog" is listed as a bad example^[4].

Style Weight (--sw) sets how hard the reference pulls. Range is 0 to 1000, default 100, and it does not work with Moodboards^[4].

Style Reference Codes point at styles in Midjourney's internal library through --sref <code>. You can mix multiple codes, combine codes with images, and use --sref random to get a random one, which converts to a fixed code on submit. Rerolls and variations keep the code, so a random hit can be locked and reused^[4]. The Style Explorer under the Styles tab on the Explore page lets you browse and search these by term.

Stylize (--s) decides how much of Midjourney's own taste overrides your literal wording. Default is 100 on a 0 to 1000 scale^[5]. Low values track the prompt, high values give the model room and drift from the exact description.

A workable pattern for series work: fix the style with --sref, hold --sw near the default until the look is right, keep --s low while you iterate wording, then raise it once the composition is settled. On clipdance.ai/midjourney these arrive as ordinary form fields rather than flags typed at the end of a prompt, which mostly matters because it stops the silent failures that come from a stray comma inside a parameter.

HD, GPU minutes, and the edit downgrade

V8 introduced HD images, and neither V6 nor V7 has the feature^[1]. HD renders at 2048px without an upscale step, toggled through the settings panel or the --hd and --sd parameters. The cost is explicit: 1.3 minutes of GPU time for HD against 0.8 for SD^[1].

Two limits come attached, and both are easy to hit.

Aspect ratio caps at 14:1 on V8 generally, but HD tops out at 4:1^[1]. A 21:9 banner is fine. A 14:1 ribbon is not, once HD is on.

More awkward, running any inpainting or outpainting tool on an HD image downscales the result to SD. Pan, Zoom Out, and Edit or Vary Region all trigger it, and recovering the resolution means running an upscale afterwards^[1]. So the sequence matters: compose and edit at SD, switch to HD for the final render. Going the other way costs you a round trip.

Where Midjourney V8.2 earns its keep

Midjourney trades literal prompt obedience for aesthetic range, and V8.2 pushes further toward the bold end of that trade by its own description^[2]. That makes it a poor fit for images that must contain readable text or follow an instruction exactly, and a strong fit for four categories of work.

Concept art and world-building. One prompt returns a grid, and raising Chaos widens how far those images land from each other. That turns a single description into several directions to evaluate instead of one answer to accept or reject.

Editorial and campaign art. Flat colour, hard shadow and a dominant hue are close to the model's resting state. When a client already has a visual language, a style reference plus a low stylize value keeps you inside it.

Character sheets and cast consistency. With --cref retired, the working method is a style reference plus a reused seed, which fixes the rendering language while the figure changes. Expect near-matches rather than exact ones, since V8 seeds are documented as 99% identical^[1].

Stills that become video. A Midjourney frame is a strong starting point for an image-to-video pass, because the aesthetic decisions are already locked in the still. Generate the frame, pick the take that holds up, then animate it through clipdance.ai/image-to-video.

The version choice inside V8 is genuine rather than cosmetic. V8.1 is documented as the faster model with tighter prompt reading and better retention of small details^[1]. Midjourney V8.2 is the one tuned for aesthetics and Personalization. Prompt-faithful production work has a real argument for staying on V8.1; work where the look carries the piece has a real argument for V8.2.

Picking a version and getting on with it

The short version: Midjourney V8.2 is a taste-and-consistency release, not a capability release. It gives you a better-behaved model and a smarter Personalization system, and it inherits a V8 feature set that dropped Quality, Turbo, Multi-Prompting, Draft Mode, Niji and the entire character-reference family. Anyone carrying prompt templates from a V6 or V7 workflow should audit them for dead flags before blaming the model for a change in output.

Three checks worth running before your next batch. Confirm no --q, --turbo, --cref or :: weights survive in your saved prompts. Decide whether your Personalization profiles live on V7 or V8, because you cannot have both directions. Sequence your edits before your HD render rather than after. Midjourney V8.2 rewards that kind of housekeeping more than it rewards longer prompts, and you can work through all three on clipdance.ai without a Discord account.

FAQ

When was Midjourney V8.2 released?

Midjourney V8.2 became the default version on July 24, 2026, with the announcement posted the following day^[1][2].

What is the difference between Midjourney V8.1 and V8.2?

V8.1 is documented as the fastest model in the lineup, rendering standard jobs about four to five times faster than earlier versions while reading prompts tightly and holding small details. Midjourney V8.2 focuses on aesthetics, image quality and Personalization^[1]. They share the same feature compatibility column, so the choice is about output character rather than available controls.

Does --cref work in Midjourney V8.2?

No. Character Reference and Character Weight are marked unsupported on V8.1 and V8.2, and were already gone in V7^[1]. Consistency on V8 comes from Style Reference, Style Weight and Seed instead.

Can I use Omni Reference with V8.2?

You can attach one, but the chart marks Omni Reference as "Uses V7" for V8.1 and V8.2^[1]. The job runs on V7, so the output is a V7 render rather than a V8.2 one.

Does the quality parameter still work in V8?

No. --q accepted values on V6 and V7 and is unsupported on V8.1 and V8.2^[1]. Leaving it in a prompt does nothing.

What does HD cost compared to SD in Midjourney V8?

HD renders at 2048px for 1.3 minutes of GPU time, against 0.8 minutes for SD^[1]. HD also caps aspect ratio at 4:1 rather than 14:1.

Do my V7 Personalization profiles work on V8.2?

Your Global V7 profile does, and it is the profile V8.1 and V8.2 use, since there is no Global V8 profile yet. Profiles you create for V8.x will not work in V7^[3].

What stylize value should I start with on Midjourney V8.2?

The default is 100, and the range is 0 to 1000^[5]. Lower values keep the render closer to your wording; higher values give the model more interpretive room and let it drift from the exact description.

References

1. Midjourney. Version: feature compatibility and comparison chart. Retrieved August 2026 from docs.midjourney.com/hc/en-us/articles/32199405667853-Version
2. Midjourney. Version 8.2 announcement. Posted July 25, 2026. Retrieved August 2026 from midjourney.com/updates/version-8-2
3. Midjourney. Personalization. Retrieved August 2026 from docs.midjourney.com/hc/en-us/articles/32433330574221-Personalization
4. Midjourney. Style Reference. Retrieved August 2026 from docs.midjourney.com/hc/en-us/articles/32180011136653-Style-Reference
5. Midjourney. Stylize. Retrieved August 2026 from docs.midjourney.com/hc/en-us/articles/32196176868109-Stylize

Further reading

• Midjourney. Parameter List. docs.midjourney.com/hc/en-us/articles/32859204029709-Parameter-List
• Midjourney. GPU Speed (Fast, Relax, Turbo). docs.midjourney.com/hc/en-us/articles/32016412137741-GPU-Speed-Fast-Relax-Turbo