Midjourney V8.1 AI Image Generator

Every image below comes from nine runs made here on 19 August 2026 — each a first take, no rerolls, with the exact prompt shown. Fast runs returned four images in 13–23 seconds; the one failure we found is shown further down, not hidden.

One prompt, four finished directions

This is the trade Midjourney has always offered, and here it survives the API: a single run returns four separate full-resolution images, not one picture and not a small 2×2 grid you still have to upscale. The four below are the complete, unedited output of one 13-second run. Read them left to right and watch the model re-decide the time of day, the sky and the palette while keeping the brief — cottage, cliff, storm, one lit window — intact in every frame.

Take 1 of 4 · Standard · 3:2

Take 2 of 4 · Standard · 3:2

Take 3 of 4 · Standard · 3:2

Take 4 of 4 · Standard · 3:2

All four images are the complete output of a single 13-second run — same prompt, no reroll, nothing discarded. One flat price covered all four.

The painterly finish people mean by “the Midjourney look”

Reviewers keep reaching for the same words about V8.1 — cinematic portraits, fantasy environments, mood — and our runs bear it out. None of the prompts below name a palette beyond two colours or give any composition instructions; the restraint, the negative space and the colour discipline are the model's own. That is the reason to pick Midjourney over a prompt-literal model when the brief is a feeling rather than a spec.

Standard · 2:3 · from a 22-second run

Standard · 2:3 · from a 15-second run

Standard · 2:3 · another take from the same run

The two fox prints are takes from one run — the composition changes, the art direction holds. The fisherman is the same model in photographic mode, one prompt later.

Draft mode: a 24-image contact sheet for the price of one run

Draft is the exploration tier: instead of four finished images you get twenty-four 512-pixel sketches from one prompt, at the same flat price as a standard run. Our test run delivered all 24 in 24 seconds. The sketches are small but surprisingly resolved — enough to judge composition, palette and mood — so the workflow is: scout a direction in Draft, then re-run the keeper in Fast at full resolution. Draft is Standard quality only; it cannot be combined with HD.

Draft · 512px · 3 of the 24 sketches one run returned

Draft · 512px · same run

Draft · 512px · same run

Three of the twenty-four sketches from a single Draft run, shown at actual quality. The full sheet took 24 seconds and cost the same as one Fast run.

Standard or HD — pixels, not polish

The quality toggle is simple and honest: Standard renders at roughly 1.2 megapixels (1344×896 at 3:2 in our runs), HD at four times that (2688×1792). Both takes below are first runs of the same macro prompt. What HD buys is pixel dimensions for print and crop headroom — not a different level of artistry, and not the same picture enlarged: Midjourney re-interprets the prompt on every run, so the two dragonflies differ.

1344×896 · 14 seconds

2688×1792 · 21 seconds

Same prompt, both first takes. Because the model re-rolls composition every run, HD is a decision to make before generating — there is no way to re-render an existing take at higher resolution here.

Keep working a result after the run

A finished run is not a dead end. From any of the four results you can request a variation (same idea, repainted), a remix (same image, new prompt), background removal, or a retexture — each a one-click operation on the history card, each returning four new takes. Below, the first take of the lighthouse run and the strong variation we made from it 18 seconds later: house, cliff and lit window survive, the sky and sea get repainted.

18 seconds

The variation was generated from the finished image on the left, not from the text prompt — that is what makes it hold the composition while repainting the weather.

• Illustrators and concept artists

  The fox prints and the cliff library are one-line prompts with no composition instructions. If your work trades on mood and finish rather than spec-sheet accuracy, this is the model whose defaults are already art-directed.

• Art directors building mood boards

  Four directions per prompt in Fast, twenty-four in Draft. One brief becomes a wall of options in under a minute — then variation and remix turn the keeper into a family of frames.

• Occasional Midjourney users priced out of a subscription

  The official service starts at a monthly plan whether you generate three images or three thousand. Here a run is a fixed credit price, and a month where you create nothing costs nothing extra.

• Anyone who wants Midjourney inside a bigger toolkit

  The same workspace and subscription cover GPT Image 2, Seedream, Nano Banana, Kling and Veo — draft an idea here, restyle or animate it with a sibling model without moving files between services.

• Not for you: text-heavy design work

  Our menu-board test failed in all four takes. If the deliverable is a poster, label or UI mockup where words must be spelled correctly, start with GPT Image 2 instead — its page shows the same kind of test passing.

Credits, reference-image limits, ratio counts and resolution tiers are read from this workspace's own model registry on 19 August 2026 — they describe what you get here, not the vendors' published specifications. Midjourney and GPT Image 2 timings are our own measurements on the test sets shown on each model's page; cells we have not benchmarked say so rather than borrowing a vendor number.