Midjourney V8.2 tests --preview and moodboards in creator runs

• Midjourney quietly opened V8.2 testing through the --preview flag, and Midjourney's update post framed it as an early look at new aesthetics and personalization.
• The strongest early creator pattern is not one house style but range: AllarHaltsonen's first runs, lloydcreates' water-and-helmet set, and gizakdag's surreal portraits all push in different directions.
• Midjourney is also pairing the preview with workflow changes, because Midjourney's update post says big batch draft mode now works with --sref random, which it claims explores style space 24 times faster.
• In replies, Midjourney's moodboards reply explicitly told testers to try V8.2 with personalization and mood boards, while AllarHaltsonen's moodboards post showed why creators latched onto that combination.
• The preview still looks rough around the edges: lloydcreates' setup reply said the product was still on 8.1 with 8.2 exposed as a preview, and _VVSVS's ultrasound complaint showed at least one prompt falling into a bizarre failure mode.

You can see the official switch in Midjourney's update post, the moodboard nudge in Midjourney's reply about mood boards, and the hands-on recipe in lloydcreates' setup reply. The fun part is how quickly creators split into different test tracks: gizakdag went straight for body-horror portraiture, lloydcreates mixed --preview with Figma shaders, and chrisfirst's timeline demo pointed at prompt-to-prompt sequencing on top.

The official entry point is tiny. Midjourney's update post told users to add --preview for an early peek at V8.2 aesthetics and personalization, and Midjourney's live-now reply confirmed the test was already active.

That matches how creators actually got in. gizakdag's reply about the flag reduced the whole rollout to "Just add --preview to your prompt," while ai_artworkgen's early test post said David mentioned the flag in Midjourney Office Hours before any broader formal release language showed up in the evidence.

Midjourney's own reply language points testers toward a specific combo: Midjourney's reply about mood boards asked people to try V8.2 with personalization and mood boards.

Creator replies quickly filled in the practical stack. According to lloydcreates' setup reply, the useful combinations were:

• Add --preview to access the 8.2 model.
• Combine --preview with --draft to get draft runs on the new model.
• Vary those drafts into high-res versions afterward.
• Pair the flow with --sref random for broad style exploration.
• Avoid combining it with a fixed style reference if you want that randomization to work.
• Mix moodboards and profiles to feel out what the preview is doing.

The evidence here is more workflow than manifesto, which is why this feels like Christmas come early for Midjourney tinkerers.

The other half of the ship is speed. In the same post, Midjourney's update post said big batch draft mode now works with --sref random, producing 24-image style sweeps and claiming 24x faster exploration.

That mechanic shows up directly in the attachments.

The practical implication is simple: V8.2 testing is happening inside a wider style-search workflow, not as one precious prompt at a time. That is why so many of the independent posts look less like polished hero shots and more like sampling passes.

The early V8.2 gallery is weirdly broad. AllarHaltsonen's first runs leaned into ad-like object rendering and branded fashion images, lloydcreates' water-and-helmet set pushed glossy sci-fi minimalism, and gizakdag's surreal portraits went full anatomical nightmare.

A few recurring visual behaviors show up across those runs:

• Cleaner studio backgrounds, in lloydcreates' portrait tests and AllarHaltsonen's first runs.
• Strong motion and smear effects, in AllarHaltsonen's motion series and lloydcreates' soccer combat set.
• More aggressive surreal facial deformation, in gizakdag's surreal portraits.
• High-gloss metallic and glass surfaces, in lloydcreates' water-and-helmet set.
• Better cohesion across a themed set, in AllarHaltsonen's horses-and-supercars series.

That range is the real story. The preview is not being praised for one default look, but for how many different looks creators are already able to pull out of it.

Some of the more interesting tests are happening one layer above raw prompting. lloydcreates' soccer combat set combined V8.2 preview outputs with Figma shaders, while chrisfirst's timeline demo showed a promotable timeline interface that morphs results across a sequence of prompts.

That matters because the preview is being used as source material for downstream motion, compositing, and interface experiments, not just for static image drops. The creator behavior in this evidence pool looks closer to pipeline building than model gawking.

The preview is not being framed as finished, even by Midjourney. Midjourney's still-working-on-it reply told one tester the team was still working on it, and lloydcreates' setup reply described 8.2 as a preview while saying the main product was still on 8.1.

The oddest concrete miss came from _VVSVS's ultrasound complaint, who said the new model kept generating ultrasound imagery. That one post does not define the model, but it does add a useful counterweight to the flood of celebratory example grids.