Midjourney V8.2 is the version Midjourney serves by default, and using it starts with two constraints the official pages state plainly: there is no free tier, and there are only two interfaces.
What Midjourney V8.2 is
Midjourney’s version document carries the sentence that matters most: “The current default Midjourney version is V8.2.” It became the default on 24 July 2026. The release notes describe a focus on aesthetics, image quality and personalization, with images that should be “more creative, bold, sophisticated, edgy and fresh”.
One structural change is easy to miss. The same page says V8.2 “features the new Edit Model, which replaces Omni Reference, Character Reference, and the Retexture tool.” Tutorials built on --cref, --cw or --oref describe an earlier pipeline: here you attach reference images to the prompt, up to four at once, and edit with instructions.
Access is the second constraint. Midjourney runs from its website or from Discord, nowhere else. On 16 July 2025 the company said it was “starting to investigate opening up an Enterprise API”, and no endpoint, key or price has appeared since. The Midjourney API guide documents that boundary, including the terms-of-service clause that forbids automated tools.
Where Midjourney V8.2 runs
| Surface | What the official documentation attaches to it |
|---|---|
| Website | The Create page and its Imagine bar; pressing Enter returns a set of four images |
| Discord | The Midjourney Bot answers with four options inside a minute; /info reports fast time remaining |
| Edit Model | Instruction-based editing, image references for generation, inpainting and outpainting |
| Enterprise API | Described as under investigation; no endpoint, key or price published |
Every generation counts against the GPU time bundled with the subscription, on either surface.
The Midjourney V8.2 hub gathers the same official figures in one place, with the two sample images from the release notes.
Your first Midjourney V8.2 image, step by step
- Subscribe first, because the Getting Started Guide opens with that requirement.
- Open the Create page at midjourney.com/imagine and find the Imagine bar.
- Write one short sentence: subject, what it is doing, the setting.
- Press Enter. A set of four images comes back.
- Change one thing and run it again. A controlled edit teaches more than a fresh prompt.
The first grid is a reading of your prompt, not a finished asset.
Worked example: a copyable poster prompt
A poster for a small ceramics studio, portrait format, with no line of type in the prompt — the official guidance is to keep prompts short. Case one:
a potter's hands turning a wide bowl on a kick wheel, clay dust in the air, single window light from the left, muted earthy palette, calm mood, close three-quarter composition --ar 2:3No reference image here. On default settings four variations return: the bowl curves slightly differently in each frame, and the hands stay plausible across all four. At 2:3 a standard render is 896 × 1344 pixels, enough for a proof and not for print; the Creative and Subtle upscalers reach 1792 × 2688.
Case two changes one block and leaves the rest alone: “same potter’s hands, same bowl, same window light; replace the workshop with a bare concrete wall and morning light from a high window; keep the composition and the palette.” Holding the hands still across two runs shows which words carry the subject and which carry the scene.
Prompt structure: short beats long
Midjourney’s own guidance is blunter than most tutorials: “Short and simple prompts typically generate the best images with Midjourney. Use phrases that clearly describe what you want to see… Avoid making long lists or detailed instructions; these can confuse the process.”
The seven elements Midjourney lists
The official Prompt Basics page names the elements worth covering and stops there: subject, medium, environment, lighting, color, mood and composition. Filling all seven in one sentence is already long by Midjourney’s standard, and the page states the trade-off directly: “Fewer details mean more variety, but you get less control over the outcome.”
Three rules for parameter formatting
Three rules in the official parameter list break a prompt quietly: keep parameters at the end; keep a space between the prompt and the first double dash; use no punctuation inside a parameter value.
Two further facts decide whether a prompt runs as written. The default aspect ratio is 1:1, and --ar cannot take decimals — write 139:100, not 1.39:1.
Resolution, HD and GPU minutes in V8.2
V8.2 does not render every aspect ratio at the same pixel count, so the resolution question has a table for an answer.
| Aspect ratio | Standard output | HD output |
|---|---|---|
| 1:1 | 1024 × 1024 | 2048 × 2048 |
| 4:3 | 1232 × 928 | 2464 × 1856 |
| 2:3 | 896 × 1344 | 1792 × 2688 |
| 16:9 | 1456 × 816 | 2912 × 1632 |
Two footnotes belong with it: exact HD pixels may vary slightly at some ratios, and HD images generate at 2048 pixels and cannot currently be upscaled further.
Compute is metered in GPU minutes — 0.8 for a standard render against 1.3 for HD, a 62.5% uplift. Upscaling costs extra too, and the documentation warns it “can cost twice as many GPU minutes as generating your initial images.” Standard mode accepts ratios up to 14:1; HD is capped at 4:1.
Plans, rights and what Midjourney has not published
Four tiers, metered in GPU minutes
| Plan | Monthly | Annual | Fast GPU time | Relax and Stealth |
|---|---|---|---|---|
| Basic | $10 | $96 | 200 minutes | No Relax, no Stealth |
| Standard | $30 | $288 | 15 hours | Unlimited Relax images |
| Pro | $60 | $576 | 30 hours | Plus Stealth and Relax video |
| Mega | $120 | $1,152 | 60 hours | Plus Stealth and Relax video |
Extra compute costs $4 per GPU hour, and paying annually takes 20% off. On rights, Midjourney states that you own what you create, with one condition: a company, or an employee of one, earning more than $1,000,000 a year must be on Pro or Mega to own it.
The unpublished side of the ledger matters as much. No free tier: the subscription document lists four paid plans and nothing else. No public API. Training data, parameter count, architecture and official scores are all undocumented, and percentages circulating online come from third-party sites.
Where licensing decides your tool rather than the interface, Adobe Firefly Image documents a licensed-training position and Stable Diffusion the downloadable-weights route. Recraft is the neighbouring topic for design assets.
Frequently asked questions
Can I try Midjourney before paying? The subscription page lists four paid tiers and no free option, and the Getting Started Guide opens with the requirement to buy one.
Does Midjourney have an API? No. The only official statement is that an Enterprise API is under investigation, and the terms of service forbid automated tools — so a service advertising programmatic access is automating a consumer interface.
How many images come back per prompt? Four, on the website and in Discord alike, each counting against the GPU time included with your plan.
Which plan do I need for Stealth Mode? Pro or Mega. Unlimited Relax image generation starts at Standard, and unlimited Relax video starts at Pro.
Why does a longer prompt give worse results? Midjourney’s own guidance is that short, clearly phrased prompts work best and long lists confuse the process. Fewer details buy variety; more buy control.