Images
An overview of Midjourney's models

Key takeaways
V8.2 has been Midjourney's default image model since July 24, 2026. V8.1 held the default from June 10 to July 23, and V7 held it for about a year before that.
You switch versions by adding "--v" and a number to the end of a prompt, by picking a default in the settings panel, or by using "--niji 7" for the anime line.
V8.1 and V8.2 add HD images at 2048 pixels, a 24-image Draft mode, and the Edit Model. They don't support Omni Reference, the quality parameter, or Turbo mode.
Niji 7 is a separate anime-focused model that launched January 9, 2026. The compatibility chart in the docs lists no Niji version for V8.1 or V8.2.
Midjourney's video model animates a start image into a 5-second clip and can extend it to 21 seconds.
Which Midjourney models are current?
Midjourney's default model is V8.2, released July 24, 2026. The Version docs cover V8.1, V7, V6.1, V6, and Niji 7, and the video model is a separate tool that starts from an image.
No new numbered version has shipped in the last 60 days. The newest items on Midjourney's updates page, dated September 24 to October 8, 2026, cover the Edit Model and the alpha website.
Model | Released | Default status | Documented focus |
|---|---|---|---|
V8.2 | July 24, 2026 | Current default | Aesthetics, image quality, Personalization, Edit Model |
V8.1 | April 14, 2026 | Default June 10 to July 23, 2026 | Speed, prompt reading, small details, HD |
V7 | April 3, 2025 | Default June 17, 2025 to June 9, 2026 | Prompt precision, textures, Draft mode, Omni Reference |
V6.1 | July 30, 2024 | Default until June 16, 2025 | Coherence, details, about 25% faster than V6 |
V6 | December 20, 2023 | Default February 14 to July 30, 2024 | Longer prompts, image prompting, remixing |
Niji 7 | January 9, 2026 | Separate anime line | Line work, coherence, a flatter look |
Version history in brief
Midjourney's docs go back to Version 4, released in November 2022 on a new codebase. Version 5 followed in March 2023, with 5.1 in May and 5.2 in June. V6 arrived on December 20, 2023, and V6.1 on July 30, 2024.
V7 came out April 3, 2025 and became the default in mid-June 2025. It introduced Draft mode and Omni Reference. Midjourney then ran an early-access V8 alpha from March 17 to July 24, 2026. Images made in it stay in your gallery, but the version itself is retired. V8.1 followed on April 14, 2026, and V8.2 replaced it as the default on July 24.
How to switch between versions
Parameters go at the end of a prompt, after the description, with a space before the dashes and no punctuation inside them. The docs list these for picking a model:
Version: add "--v 7" or "--v 8.2" to the end of a prompt in the Imagine bar or in Discord.
Default version: click the settings button in the Imagine bar, or run the settings command in Discord, to set one version for every future prompt.
Niji: add "--niji 7" in Discord, or choose Niji 7 from the Version dropdown on the web.
Resolution: "--hd" and "--sd" switch V8.1 and V8.2 between 2048-pixel and 1024-pixel images.
Edit: "--edit" followed by an image URL runs the Edit Model in Discord.
What V8.1 and V8.2 do and where they stop
Midjourney calls V8.1 its fastest model so far, with standard jobs rendering about 4 to 5 times faster than earlier versions, and says it reads prompts better and holds on to small details. Raw mode strips default styling for closer prompt adherence. V8.2 then focused on aesthetics, image quality, and Personalization, and the docs say its images are "more creative, bold, sophisticated, and edgy."
HD renders about 2K images without upscaling and costs 1.3 minutes of GPU time against 0.8 for standard. The maximum aspect ratio is 14:1, or 4:1 for HD. Seeds on V8.1 and V8.2 give results the chart calls "99% identical."
The documented limits are mostly features these versions leave out. The chart shows no Omni Reference, no quality parameter, no multi-prompting, and no Turbo mode. HD also has a catch, since inpainting and outpainting tools downscale HD images to standard size until you run an Upscale option.
The Edit Model is the biggest change in this generation. It works on V8.1 and V8.2 and replaces Omni Reference, Character Reference, and the Retexture tool. It edits an image from written instructions, builds new images from up to four reference images, and handles inpainting and outpainting in the Editor. It isn't compatible with "--tile" or Remix, and the docs suggest merging characters into one reference image if their features start to blend.
What older versions still handle
Some features exist only on earlier models. Omni Reference, with its weight setting, works on V7 and not on V8. The quality parameter takes 1, 2, or 4 on V7 and 0.5, 1, or 2 on V6. Multi-prompting appears only in V6. Retexture works on versions 7 and earlier, and Character Reference runs on V6 and Niji 6.
V7's Draft mode makes 4 images at half the GPU cost of a standard V7 prompt.
Niji 7 and the anime line
Midjourney describes Niji models as a special series focused on Eastern and anime aesthetics and illustrative styles, with their own website and Discord server. Niji 7 launched January 9, 2026 with a boost in coherency, so fine details like eyes, reflections, and small background elements render more clearly. It follows prompts more closely too.
The docs warn that this version is more literal, so broad or "vibey" prompts may not behave the same as before. Niji 7 introduced a cleaner, flatter look meant to show off its line work. The Niji 7 announcement adds that character reference "won't be supported by niji 7" and that style references carry over. Niji 6, released June 7, 2024, appears in the chart as "--niji 6" on the V6 side.
The video model
Midjourney announced its V1 video model on June 18, 2025. According to the video docs, you feed it an image as the first frame, add an optional text prompt, and get a 5-second clip. A batch makes 4 videos by default, and you can set it to 1 or 2.
Each clip can be extended by 4 seconds up to 4 times, reaching a 21-second maximum. Motion defaults to low, which tends toward still scenes and subtle movement, while high motion allows big camera moves and may produce glitchy movement. The "--loop" and "--end" parameters set a looping or custom final frame.
Videos render at 480p, with 720p on the Standard, Pro, and Mega plans. A batch of four costs 8 minutes of GPU time at SD and 26 at HD. Style references, image prompts, and Edit Model references don't work with video.
Draft mode and GPU speeds
Draft mode trades resolution for speed. On V8.1 and V8.2, one Draft prompt returns a batch of 24 images at 512 by 512 pixels for 0.4 minutes of GPU time, and it runs only on midjourney.com. The Draft mode docs say to use Vary or Remix on a favorite to regenerate it at full resolution. Conversational mode, where an AI writes prompts from plain language, works with V7 and V8.1.
Job | Approximate GPU cost |
|---|---|
SD image | 0.8 minutes |
HD image | 1.3 minutes |
Edit Model SD image | 1 minute |
Edit Model HD image | 2.3 minutes |
Draft batch (V8.1 and V8.2) | 0.4 minutes |
Relax mode is unlimited for images on Standard, Pro, and Mega, with waits of 0 to 30 minutes. Turbo mode can run up to 4 times faster than Fast at double the cost, and the docs say it isn't supported in V8.1.
Personalization and style references across versions
Personalization works on versions 6 and later. Midjourney says your V7 Global Profile works in V8.1 and V8.2, there's no Global V8 profile, and V8 profiles aren't compatible with V7. The stylize parameter sets how much of a profile applies, from 0 to 1000, with a default of 100.
Style references work on version 6 and later. They capture colors, medium, textures, and lighting but not objects or people. Style weight runs from 0 to 1000 with a default of 100, and numeric style codes or "--sref random" pull from Midjourney's style library. With the Edit Model, the docs suggest describing the style in your prompt to activate it more consistently.
Plans and access
The plans page lists four subscriptions at $10, $30, $60, and $120 a month, with a 20% discount for annual billing. Standard and above get unlimited Relax images, Pro and Mega add Relax video at SD, and Stealth mode is limited to Pro and Mega.
All four plans fall under general commercial terms, but a company with more than $1,000,000 in annual gross revenue must buy Pro or Mega. Who owns a generated image is a different question, which our copyright explainer covers.
Recent changes to the Edit Model and alpha site
Midjourney opened its first V8.2 Edit Model to everyone for testing on August 27, 2026, and later updated its image quality. A September 24 update made targeted edits change only the selected pixels, which allows repeated edits without degrading the image, and made tiled images on V8.1 and V8.2 blend without visible seams.
On October 7 and 8, Midjourney began testing a Thinking Mode on its alpha website. It reruns V8.2 jobs by looking for where the image missed the prompt, such as objects, layout, anatomy, or text, and generating again. Midjourney calls it an early version.
Midjourney models FAQ
What is Midjourney's default version?
V8.2. Midjourney released it as the default on July 24, 2026, after V8.1 held the role from June 10 to July 23.
How do I change the version?
Add "--v" and the number to the end of a prompt, or set a default in the settings panel on the web or with the settings command in Discord. For anime, use "--niji 7".
Does V8.2 support Omni Reference or Character Reference?
No. The Edit Model replaces both on V8.1 and V8.2 and takes up to four reference images. Omni Reference still works on V7.
Does Midjourney have a video model?
Yes. It animates a start image into a 5-second clip, extendable to 21 seconds, at 480p or 720p depending on your plan.
Can I still use older versions?
The docs still document V7, V6.1, and V6, and the compatibility chart shows which features each supports. V8.0, the alpha, is no longer available for use.
