Midjourney has quietly swapped its default engine twice in 2026. V8.1 first took the automatic slot on June 10, 2026, capping off roughly a six-week alpha that began April 30, 2026, and then Midjourney handed that same default slot to V8.2 on July 24, 2026, again with no announcement banner and no opt-in required. Anyone who opened Discord and typed /imagine during either transition got noticeably different output than the V7 renders they were used to. If your prompts suddenly look different, or you’re setting up Midjourney for the first time, this walkthrough covers the full process built around V8.1’s still-current toolset: account setup, plan selection, the new HD mode, personalization, Omni-Reference, and a complete branded project built from scratch.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Is Midjourney V8.1 (And What Changed From V7)
Midjourney V7 launched in alpha on April 3, 2025, and became the platform’s default model on June 17, 2025, according to Midjourney’s own release notes. It was the first Midjourney model that turned personalization on by default, and it introduced Omni-Reference, a way to feed the model a reference image so a character or object stays consistent across generations. V7 held the default spot for well over a year, per Midjourney’s version history on Wikipedia.
V8 arrived in alpha on March 17, 2026, rebuilt on a GPU-native codebase that 2026 coverage pegs at roughly 5x faster than the prior generation in HD mode. V8.1 followed in alpha on April 30, 2026, and after about six weeks of testing became the new default on June 10, 2026, retiring V7 from the automatic slot. That run was short: V8.2 shipped July 24, 2026 and immediately took over the default position from V8.1. A release tracker summarized the change as bringing smarter prompt handling, better text rendering, faster generation, and higher-resolution HD output, while keeping style references, personalization, and aesthetics consistent with what V7 users already knew.
Three things matter most if you’re coming from V7:
- Speed: V8’s GPU-native pipeline processes HD jobs several times faster than the old codebase, which stretches your Fast GPU hours further.
- Resolution: the
--hdflag now targets native 2K output instead of the older upscaling path. - Text rendering: in-image text (posters, signage, logos) comes out cleaner and closer to what you actually typed in the prompt.
Everything you already knew about style references, Omni-Reference, and personalization still applies. You are not learning a new tool, you’re adapting an existing workflow to a faster, higher-resolution engine.
The GPU-native codebase behind V8 and V8.1 matters beyond raw speed. Earlier Midjourney versions ran on a rendering pipeline built for an older generation of hardware, then patched repeatedly to support newer features. V8’s rebuild lets Midjourney allocate GPU memory more efficiently per job, which is part of why HD mode can now target native 2K resolution instead of relying on a separate upscaling pass bolted on after generation. In practice, that shows up as fewer visible artifacts around fine detail, sharper edges on text and typography, and generation times that stay short even at higher output resolution.
Prerequisites: What You Need Before You Start
Midjourney has no free trial as of September 2026 and no standalone desktop app for generation, so gather these before you begin:
- A Discord account (free, discord.com), since Midjourney’s primary generation interface still runs through Discord slash commands and DMs with the Midjourney bot.
- A paid Midjourney subscription: Basic at $10/month, Standard at $30/month, Pro at $60/month, or Mega at $120/month. There is no free tier.
- A credit or debit card, or PayPal, for billing through midjourney.com.
- Reference images (optional but recommended), ideally square or near-square JPG/PNG files, if you plan to use Omni-Reference or style references.
- A web browser. The midjourney.com website also supports prompting outside Discord for subscribers.
- Roughly 90 minutes to complete every step below, including account setup, your first ten generations, and the closing project.
You do not need any coding experience. Midjourney has no official public API as of September 2026, so Discord slash commands and the web app are the only sanctioned ways to generate images, and nothing here requires a terminal.
Midjourney’s Web App vs. Discord: Choosing Your Interface
Every paid subscriber can generate through either surface, and both run the same V8.1 model with the same parameter flags, so the choice comes down to workflow preference rather than capability. The web app at midjourney.com presents your prompt history as a scrollable, searchable grid, with filters for favorites, upscales, and specific projects, which makes it easier to manage a large image library over weeks or months of work.
Discord, by contrast, is better suited to fast, conversational iteration, especially if you’re collaborating with a small team in a shared channel and want everyone to see prompts and results as they happen in real time. Discord also supports slash-command autocomplete for parameters, which can be faster once you’ve memorized the flags from Step 10. This tutorial uses Discord syntax throughout because it’s the more widely documented interface, but every prompt and flag combination shown works identically if you type it into the web app’s Create page instead.
Step 1: Create a Discord Account and Join the Midjourney Server
Go to discord.com and sign up with an email address if you don’t already have an account. Verify your email, then visit midjourney.com and click the sign-in option, which links your Discord identity to a Midjourney account. Midjourney no longer requires you to join its public Discord server to generate images, since subscribers can prompt directly from midjourney.com or from a private server they control, but joining the official server is still useful for release announcements and the showcase channels where other users post prompts alongside their results.
Account creation itself takes under five minutes. Discord asks for an email, a username, a password, and a birth date for age verification. Midjourney’s own sign-up flow then asks you to authorize the Discord connection, choose a display name for your generation library, and confirm you agree to its community guidelines, which cover acceptable content and prohibit generating misleading images of real public figures.
If you do use Discord as your generation surface, you can either prompt in a public server’s designated “newbies” channel or, better, add the Midjourney bot to a private server you own so your generations don’t scroll away in a shared feed. Open your server settings, go to Integrations, and add the Midjourney bot from midjourney.com’s account page.
Step 2: Choose a Subscription Plan
Midjourney bills every plan around two kinds of generation time: Fast GPU hours, which render almost instantly and reset each month without rolling over, and Relax mode, a queued, slower path that doesn’t draw down your Fast hours. Extra Fast GPU time can be purchased on any plan for about $4 per hour once your monthly allowance runs out.
| Plan | Monthly price | Annual price (effective/mo) | Fast GPU hours/month | Relax mode | Stealth mode |
|---|---|---|---|---|---|
| Basic | $10 | $96/yr ($8) | 3.3 hours | No | No |
| Standard | $30 | $288/yr ($24) | 15 hours | Unlimited | No |
| Pro | $60 | $576/yr ($48) | 30 hours | Unlimited | Yes |
| Mega | $120 | $1,152/yr ($96) | 60 hours | Unlimited | Yes |
For this tutorial, Standard is the practical floor. Basic has no Relax mode, so once your 3.3 Fast hours run dry mid-project, generation simply stops until you buy more time or the month resets. Standard’s unlimited Relax queue means you can keep working at a slower pace for free once Fast hours are gone, which matters when you’re testing prompts and burning through iterations. Pro adds Stealth mode, which keeps your generations out of the public feed, plus higher concurrent job limits, worth it if you’re working on unreleased branding or client work.
Mega exists mainly for studios and agencies running several concurrent campaigns. At 60 Fast hours a month it rarely makes sense for a solo user, since even heavy daily use on Pro’s 30 hours is hard to exhaust without running dozens of HD batches every week. If you’re unsure which tier fits, start on Standard for the first billing cycle, track how quickly you burn through Fast hours in your account dashboard, and upgrade only if you’re consistently dropping into Relax mode before the month ends.
Sign up for annual billing if you already know you’ll use Midjourney past three months. The 20% discount applies from the first payment.
Step 3: Confirm You’re Running V8.1 and Understand the Version History
Since July 24, 2026, every new Midjourney account defaults to V8.2 automatically, after V8.1 briefly held the default slot from June 10, 2026. If your account predates the July switch and you never manually pinned a version, you’re most likely running V8.2 by default today, and the flags and workflow in this guide still apply since V8.2 inherited V8.1’s parameter set. To confirm which version you’re on, type /settings in Discord or check the model selector on midjourney.com, where the active version appears at the top of the panel. If you see V7 or an older build listed, select V8.1 or V8.2 manually from the dropdown.
| Version | Release status | Date | Key change |
|---|---|---|---|
| V7 | Alpha → Default | Apr 3, 2025 → Jun 17, 2025 | Personalization on by default, Omni-Reference introduced |
| V8 | Alpha | Mar 17, 2026 | New GPU-native codebase, ~5x faster HD rendering |
| V8.1 | Alpha → Default → Superseded | Apr 30, 2026 → Jun 10, 2026 → Jul 24, 2026 | Smarter prompt handling, better text rendering, 2K HD output |
| V8.2 | Released → Default | Jul 24, 2026 | Iterative update, replaced V8.1 as the automatic default |
Notice how short V8.1’s run as default actually was: it took the automatic slot on June 10, 2026, and Midjourney handed that slot straight to V8.2 the day it shipped, July 24, 2026, rather than holding V8.2 back as an optional manual upgrade the way it once did with V8.1’s own alpha. You can still manually select V8.1 with --v 8.1 if the workflow in this guide is what you’ve built your projects around, and older versions like V7 remain available the same way for style consistency on existing work.
Step 4: Write Your First /imagine Prompt
In Discord, type /imagine in any channel where the Midjourney bot is active, then follow it with a prompt describing the image you want. On midjourney.com, there’s a text box at the top of the Create page that works the same way without the slash command.
/imagine prompt: a weathered lighthouse on a rocky New England coast at dawn, warm golden light, mist rolling over the water, cinematic photography, shot on 35mm filmPress enter and Midjourney queues the job. On a Standard plan with Fast hours available, expect a result in under a minute for a standard (non-HD) generation. V8.1’s speed gains are most visible once you switch on HD mode in Step 6. The bot returns a grid of four variations in a single image, each numbered 1 through 4.
Keep your first few prompts specific but short. Overloaded prompts with a dozen competing style descriptors tend to confuse even V8.1’s improved prompt handling. Start with a subject, a setting, a lighting condition, and one style reference, then add complexity once you see how the model interprets your baseline description.
Step 5: Read the Grid and Use the Action Buttons
Under every four-image grid, Midjourney shows two rows of buttons: U1–U4 and V1–V4, plus a redo arrow. U (Upscale) takes one of the four quadrants and renders it as a standalone, higher-resolution image. V (Vary) generates four new variations based on one quadrant while keeping its general composition. The redo arrow reruns the entire prompt with a fresh seed for four completely different results.
A typical workflow: run the initial grid, click V on whichever quadrant is closest to what you want, review the four new variations, then click U on the winner to produce your final image. This two-stage narrowing process, vary then upscale, usually gets better results than trying to nail the exact composition on your first grid.
Below the U/V buttons you’ll also find a Vary (Region) option once you upscale, which opens an in-browser editor for repainting a specific part of the image without regenerating the whole thing. It’s useful for fixing a hand, a piece of text, or a background element without losing everything else you liked.
Two more buttons appear on an upscaled image: Zoom Out and Pan. Zoom Out expands the canvas outward while inventing new content around your original image, useful when a composition feels too tight or you need extra headroom for text overlay later. Pan shifts the frame in a specific direction, generating fresh content on the side you’re panning toward while keeping the opposite edge anchored to what already exists. Both tools cost the same Fast-hour time as a standard generation and are worth testing early, since they solve compositional problems that would otherwise require starting a prompt over from scratch.
Step 6: Turn On HD Mode for 2K Output
V8’s GPU-native codebase supports native 2K resolution through the --hd flag, replacing the older interpolated upscaling that earlier versions relied on. Add it to the end of any prompt:
/imagine prompt: a weathered lighthouse on a rocky New England coast at dawn, warm golden light, mist rolling over the water, cinematic photography, shot on 35mm film --hdHD jobs take longer and draw more heavily on your Fast GPU hours than standard-resolution jobs, so reserve it for prompts you’ve already refined through a few standard-resolution rounds. A practical pattern: iterate on composition and subject matter at default resolution, then re-run your final, locked-in prompt with --hd once for the deliverable-quality version.
Because V8’s codebase is roughly five times faster at HD rendering than the pre-V8 pipeline, HD generation on V8.1 costs noticeably less Fast-hour time per image than it did under V7. Worth remembering if you were rationing HD jobs on the older model and haven’t adjusted your workflow since the switch.
Step 7: Set Up Personalization
Personalization has been on by default since V7 and carries over into V8.1 unchanged in concept. It works by asking you to compare pairs of images and pick the one you prefer, building a private aesthetic profile the model then leans on for future generations.
Run /imagine a few times normally, then type /tune in Discord (or visit the Personalization tab on midjourney.com) to start the rating process. Midjourney generates roughly 50–100 image pairs. Click through them selecting your preference each time. The more pairs you rate, the more the model’s default output leans toward your taste without you needing to describe a style in every prompt.
Once your profile exists, add --p to any prompt to apply it, or leave it off for a neutral, unpersonalized result. You can maintain multiple personalization profiles for different projects, a photorealistic profile for client photography work and a stylized-illustration profile for a separate brand, for example, and switch between them with the --p flag followed by the profile code shown in your account settings.
Step 8: Use Omni-Reference for Consistent Characters and Objects
Omni-Reference, introduced with V7 and retained in V8.1, lets you upload a reference image and have Midjourney keep that subject, a character, a product, a mascot, visually consistent across different scenes and prompts. This solves one of the oldest complaints about AI image generation: the same “person” looking like six different people across a project.
/imagine prompt: the same woman from the reference image now standing in a modern kitchen, holding a coffee mug, soft morning light --oref [uploaded-image-url] --ow 250To use it, drag your reference image into the Discord prompt box (or upload it on midjourney.com) before typing your prompt, then reference it with --oref followed by the image’s URL, which Midjourney generates automatically once you attach the file. The --ow flag controls Omni-Reference weight from 1–1000. Higher values pull the output closer to the reference image’s exact appearance, lower values give the model more freedom to reinterpret the subject in a new pose or setting. Start around 100–250 and adjust from there.
Step 9: Apply Style References and Style Tuning
Where Omni-Reference locks a subject, style references (--sref) lock an aesthetic, color palette, brush technique, lighting mood, independent of subject matter. Upload a reference image the same way as Step 8, then attach it with --sref instead of --oref:
/imagine prompt: a mountain cabin surrounded by pine trees at sunset --sref [uploaded-image-url] --sw 400--sw (style weight, 0–1000) controls how strongly the reference influences the output. You can also skip an uploaded image entirely and use Midjourney’s built-in style codes, numeric IDs tied to specific aesthetics that the community shares and catalogs, by passing --sref [code number] instead of a URL. This is the fastest way to get a repeatable, branded look across dozens of images without re-uploading a reference file each time.
Step 10: Master the Core Parameter Flags
Beyond --hd, --oref, and --sref, a handful of parameters cover almost everything you’ll need day to day:
/imagine prompt: a minimalist product shot of a ceramic mug on a linen backdrop --ar 4:5 --stylize 250 --chaos 15 --no text, watermark --seed 48213--ar: aspect ratio, such as 16:9 for widescreen, 4:5 for a social feed post, or 1:1 for square--stylize: how strongly Midjourney’s default aesthetic bias applies, from 0 (literal) to 1000 (highly artistic)--chaos: variation between the four grid images, from 0 (near-identical) to 100 (wildly different interpretations)--no: a comma-separated exclusion list telling the model what to leave out, such as text, watermark, or extra limbs--seed: a fixed number that, combined with an identical prompt, produces reproducible results across runs
Stack these flags in any order at the end of your prompt. Low --chaos values are useful once you’ve found a composition you like and want four close variants. High values work better for early exploration when you’re not sure what direction to take a project.
Step 11: Upscale, Vary, and Export Final Images
Once you’ve landed on a final upscaled image, right-click it in Discord (or use the download icon on midjourney.com) to save the full-resolution file. HD-mode upscales export at native 2K on the long edge, while standard upscales export smaller, roughly in the 1024–1536 pixel range depending on aspect ratio.
If you need even higher resolution for print, Midjourney doesn’t offer a built-in super-resolution pass beyond the standard 2K HD ceiling, so plan on running the file through a dedicated upscaler (Topaz Gigapixel or similar) for anything larger than a standard social or web asset. Keep your original seed number recorded alongside any file you might need to regenerate or tweak later. Without it, reproducing an exact composition from scratch is far harder even with an identical prompt.
Step 12: Batch a Prompt List for a Repeatable Workflow
Because Midjourney has no official public API, there’s no scripted way to fire off dozens of prompts programmatically without violating its terms of service, which explicitly prohibit automation. That puts it in different territory from API-first tools like FLUX.2 or the GPT Image 2 API, both built for scripted batch generation from the ground up. The practical workaround for repetitive Midjourney projects, a product catalog, a set of blog headers, a card deck, is to prepare your full prompt list in a plain text file first, then paste them into Discord one at a time in quick succession while Fast hours are available.
/imagine prompt: ceramic mug, white background, studio lighting --ar 1:1 --sref 4021837 --seed 100
/imagine prompt: ceramic bowl, white background, studio lighting --ar 1:1 --sref 4021837 --seed 101
/imagine prompt: ceramic plate, white background, studio lighting --ar 1:1 --sref 4021837 --seed 102
/imagine prompt: ceramic vase, white background, studio lighting --ar 1:1 --sref 4021837 --seed 103Notice the shared --sref code and incrementing seed values. This keeps every item in the set visually consistent while still giving each one a distinct, reproducible result. Queue four or five jobs at once. Midjourney processes them in parallel up to your plan’s concurrent job limit (three on Basic/Standard, twelve on Pro/Mega), so you’re rarely waiting on one image at a time.
Complete Working Project: A Six-Poster Brand Campaign in V8.1
Here’s the full process end to end, using everything covered above to produce a consistent six-poster series for a fictional coffee brand.
- Pick a style reference. Generate one exploratory image without
--srefto find a look you like, then upscale it and note its job ID. You can pull the style code directly from that image’s info panel for reuse. - Lock a seed for structure. Run your base poster layout prompt with
--seedset, so every subsequent flavor variation shares the same underlying composition. - Write the six prompts, one per product (espresso, cold brew, latte, drip, decaf, seasonal blend), all sharing the same
--srefcode,--ar 4:5, and a consistent--stylizevalue. - Add
--hdonly once you’ve confirmed the composition works at standard resolution. Running all six in HD from the start burns Fast hours on drafts you may throw away. - Batch-queue the final six using the prompt-list method from Step 12, pasting one after another.
- Upscale each winning grid quadrant individually, using V (vary) first if any single image needs a tweak before the final U (upscale) pass.
/imagine prompt: minimalist coffee bag poster, espresso beans scattered on dark walnut table, warm rim light, editorial product photography --ar 4:5 --sref 4021837 --stylize 300 --hd --seed 7001
/imagine prompt: minimalist coffee bag poster, cold brew glass with ice, warm rim light, editorial product photography --ar 4:5 --sref 4021837 --stylize 300 --hd --seed 7002Total time for a six-image set like this, including two rounds of variation per poster, typically runs 30–45 minutes on a Standard plan with Fast hours available. That’s well within the 15 hours Standard includes each month, even accounting for the earlier exploratory prompts from Steps 1–11.
Common Pitfalls When Switching From V7 to V8.1
- Assuming your V7 style codes render identically. Style reference codes carry over, but V8.1’s rendering pipeline interprets them slightly differently. Always re-test a saved
--srefcode before committing a whole batch to it. - Burning HD mode on exploratory prompts. HD costs more Fast-hour time per image, so save it for the composition you’ve already validated at standard resolution.
- Overloading prompts with stacked adjectives. Even with V8.1’s improved prompt handling, ten competing style descriptors dilute the result more than they refine it. Pick your two or three most important qualifiers.
- Forgetting
--seedon anything you might need to reproduce. Without it, an “almost identical” re-run can drift noticeably in lighting, framing, or color from the original. - Running Basic plan for a multi-day project. 3.3 Fast hours disappears fast once you’re doing multiple variation passes per image, and Basic has no Relax fallback.
- Ignoring the difference between
--orefand--sref. Using a style reference when you actually need subject consistency (or vice versa) produces images that look right in isolation but drift the moment you generate a second scene.
Troubleshooting Midjourney V8.1
- The bot isn’t responding to
/imagine. Confirm the Midjourney bot has permission to read and send messages in that specific channel. Server admins sometimes restrict bot access per-channel. - Generated text still looks garbled despite V8.1’s improvements. Keep on-image text short (under six words) and put it in quotes within the prompt, since long sentences or paragraphs still push past what any current diffusion model renders reliably.
- Omni-Reference isn’t holding the subject’s face consistent. Raise
--owtoward 400–500 and make sure your reference image shows the subject clearly, front-facing, without heavy shadow or motion blur. - Jobs are stuck in a long queue. You’ve likely exhausted your Fast GPU hours and dropped into Relax mode, which queues behind every other Relax job on the platform. Check your remaining Fast time under
/info. - Style reference produces a completely different look than expected. Lower
--swif the reference is overpowering the subject matter, or confirm you uploaded the intended image (drag-and-drop order in Discord can silently attach the wrong file). - HD mode seems slower than promised. Peak-traffic hours (US daytime, especially weekday afternoons) slow Fast-mode queues platform-wide regardless of your plan tier.
- Can’t tell which version generated an old image. Click the image in your midjourney.com library. The info panel lists the exact model version and full parameter string used.
- Personalization results feel inconsistent. Rate at least 100 pairs during
/tunebefore judging results, since profiles built on fewer than 50 ratings are noisy. - Payment declined mid-cycle. Midjourney suspends Fast-hour access immediately on failed billing. Update your card under Account Settings on midjourney.com, not through Discord.
Advanced Tips for Power Users
Combine a locked --seed with small --stylize increments (try 100, 200, 300 on the same seed and prompt) to see exactly how the stylize parameter shifts a composition without any other variable changing. This is the fastest way to learn what each flag actually does rather than guessing from documentation.
For teams sharing a Pro or Mega plan, set up a dedicated private Discord server with separate channels per project. Midjourney’s job history stays searchable by channel, which turns into a simple asset library over a few months without needing an external DAM tool for early-stage exploration work.
Use Vary (Region), available after upscaling, to fix small compositional issues instead of regenerating the whole image. Selecting just the problem area and re-prompting that region alone preserves everything else you already approved, which saves both time and Fast-hour budget compared to starting over.
When building a multi-image brand set, generate your style reference image first, upscale it, and treat that single file as a locked asset you never regenerate. Every other image in the campaign should point back to that one --sref source rather than each poster developing its own drifting interpretation of “the brand look.”
Commercial Use and Licensing on Paid Plans
All four paid tiers, Basic through Mega, grant commercial usage rights to images you generate, meaning you can use them in client work, marketing, packaging, or resale without an additional license fee. That’s a meaningful difference from tools that gate commercial rights behind a separate enterprise tier. The practical limits sit elsewhere: Stealth mode (Pro and Mega only) keeps a generation out of the public community feed, which matters for unreleased branding, but even on Standard your images are yours to use commercially the moment they’re generated, public feed visibility aside.
One caveat worth building into any client-facing workflow: because Midjourney trains on a broad mix of publicly available imagery, output can occasionally echo recognizable visual styles closely enough to raise questions in commercial contexts, particularly branded work or anything resembling a living artist’s signature style. Running a reverse-image check on final deliverables before client handoff is a reasonable habit if the campaign has any legal exposure, the same due diligence most studios already apply to stock photography and licensed illustration.
Midjourney V8.1 vs. Stable Diffusion 3.5 vs. Adobe Firefly
Midjourney V8.1 isn’t the only serious option for AI image generation in September 2026. If you’d rather compare a full field of tools before committing to one, our AI image generator roundup covers the wider market. Stability AI’s Stable Diffusion 3.5 remains its current flagship open-weight release, with no successor announced as of this writing, and it’s still described by Stability AI as its most powerful image model. The company’s 2026 launches went elsewhere, into Stable Audio 3.0 and the Brand Studio platform. Our own Stable Diffusion 3.5 setup tutorial walks through self-hosting it if that’s the direction you need. Adobe, meanwhile, retired Firefly Image 3 from Photoshop starting April 28, 2026, fully removing it from all plans by August 2026, with Firefly Image 4 and 4 Ultra now the default models inside Photoshop’s Generative Fill and Expand tools. See our Adobe Firefly walkthrough for the Creative Cloud-integrated workflow.
| Tool | Access model | Pricing model | Official API | Best for |
|---|---|---|---|---|
| Midjourney V8.1 | Closed, hosted (Discord + web) | Subscription, $10–$120/mo | None | Stylized, artistic, campaign-consistent imagery |
| Stable Diffusion 3.5 | Open-weight, self-hostable | Free (self-host) or API/credits via partners | Yes, via Stability API and partners | Local deployment, fine-tuning, developer pipelines |
| Adobe Firefly (Image 4/4 Ultra) | Closed, integrated into Creative Cloud | Credit-based, 10–20 credits per image depending on model | Yes, Firefly Services API | In-workflow editing inside Photoshop/Illustrator |
The practical split: pick Midjourney V8.1 when the deliverable is a finished, art-directed image and you’re comfortable working through Discord or the web app manually. Pick Stable Diffusion 3.5 when you need to self-host, fine-tune on your own data, or integrate generation into a pipeline through an actual API, something Midjourney still doesn’t offer. Pick Adobe Firefly when the image needs to live inside an existing Photoshop or Illustrator workflow, especially for commercial work where Adobe’s content-credentialing and licensing terms matter more than raw stylistic range.
Frequently Asked Questions
Is Midjourney V8.1 free to use?
No. Midjourney has no free trial or free tier as of September 2026. The cheapest option is the Basic plan at $10/month, which includes 3.3 Fast GPU hours and no Relax mode.
How do I switch from V7 to V8.1 manually?
Every account defaulted to V8.1 automatically on June 10, 2026, and then to V8.2 on July 24, 2026, so most accounts today are actually running V8.2 rather than V8.1. If you’re still seeing V7 output, open /settings in Discord or the model selector on midjourney.com and choose V8.1 or V8.2 from the version list, or add --v 8.1 (or --v 8.2) to the end of any prompt.
Does Midjourney V8.1 have an official API for developers?
No. As of September 2026, Midjourney offers no official public API, and its terms of service explicitly prohibit automated access. Third-party wrapper tools exist but operate outside Midjourney’s sanctioned usage terms.
What resolution does Midjourney V8.1 output at?
Standard generations render at a lower base resolution optimized for speed, while the --hd flag activates V8’s GPU-native pipeline for native 2K output on the long edge, a jump from the interpolated upscaling earlier versions used.
What’s the difference between Omni-Reference and style reference?
Omni-Reference (--oref) locks a specific subject, a character or object, so it stays recognizable across different scenes. Style reference (--sref) locks an aesthetic, like a color palette or rendering technique, independent of what subject appears in the frame.
Why did my style reference code look different after the V8.1 switch?
V8.1’s rendering pipeline interprets saved style codes and reference images slightly differently than V7 did, even though the style reference system itself carried over unchanged conceptually. Re-test any saved --sref code on a single image before committing a full batch to it.
Which Midjourney plan should I choose for a one-time project?
Standard, at $30/month, is the practical minimum for most one-off projects because it’s the cheapest tier with unlimited Relax mode. Basic’s 3.3 Fast hours run out quickly once you factor in variation and upscale passes, and Basic has no fallback once they’re gone.
Can I still generate images with V7 after the V8.1 default switch?
Yes. Older versions including V7 remain available as a manual selection via --v 7 or through the version dropdown in settings, which is useful if you have an existing project whose look depends on V7’s specific rendering characteristics.
Related Coverage
- How to Build Claude Code Skills: 12 Steps, 90 Min [2026]
- How to Use Qwen-Image-3.0: 12 Steps, 90 Min [2026]
- Build an AI Image Generator API: 13 Steps, 90 Min [2026]
- How to Use Google AlphaEvolve: 12 Steps, 90 Min [2026]
- How to Use Leonardo AI Image Generator: 12 Steps [2026]
