Midjourney V8 is live on alpha: Here's what's actually different

A practical look at MJ v8, the three style tools you should start with, and what's changed about how it works.

Mar 23, 2026

Welcome back to Visually AI!

Today’s reading time is 5 minutes.

In today’s edition:

Midjourney v8 Alpha is live
The three style tools to use
Midjourney v7 & v8 Comparison
Midjourney v8 Prompts

Midjourney V8 is live — but only on alpha

As of this week, Midjourney’s V8 model is available for community testing at alpha.midjourney.com. It’s not on the main site yet, and it’s not in Discord. You need to head to the Alpha platform specifically.

This is an early release, Midjourney are asking for ratings and feedback as they refine the model. Here’s a straightforward look at what’s changed and how to start using it:

What’s new in V8

Images generate roughly 5x faster than V7, and V8 introduces native 2K resolution via the --hd flag (no upscaling); it renders at that resolution from the start.

Prompt understanding has also improved. V8 picks up on more detail in your prompts including specific colors, lighting conditions, spatial layout and material textures. Complex compositions that V7 would partially ignore now render closer to what you described.

Text in images works much better too. Put your text in “quotes” within the prompt, and V8 handles it with higher accuracy — useful for anything with signage, typography, or labels.

Parameters currently supported: --chaos, --weird, --exp, --raw, --hd, and --q 4 (for extra coherence). Multiple aspect ratios are supported. Relax mode is now available for V8.

A note on cost: --hd and --q 4 each cost 4x your standard GPU time. Style references and moodboards have been updated — the new version (--sv 7) is 4x faster and 4x cheaper. More on that below.

The three style tools worth knowing

V8 is built to work best when you use its stylization systems. Here’s how each one works:

Style Creator

Style Creator lets you build a custom style code (--sref) by selecting images rather than writing prompts. You go to the Style Creator page, enter a prompt, and then choose from a grid of generated images — picking the ones whose visual style matches what you’re going for. The model uses your selections (and your rejections) to build a reusable style code.

It refines in rounds. Most styles stabilize after 5–10 rounds. Rounds 10–15 bring more detail. Past round 15, changes are minimal.

Once you have a code, you apply it to any prompt with --sref [your code]. You can also save codes as bookmarks and organize them into moodboards.

--sref (Style References)

If you already have an image whose style you want to replicate, you can use it directly as a style reference. Add --sref [image URL] to your prompt and V8 pulls the visual style — colour palette, texture, lighting, medium — without copying the content or people in the image.

Style codes generated in V7 carry over. Backwards compatibility is supported.

Srefs also now default to --sv 7 — 4x faster, 4x cheaper, and compatible with --hd, personalization, --stylize, and --exp. If you need the previous version, add --sv 6 to your prompt.

Moodboards

Moodboards let you combine multiple reference images into a single style input. Rather than anchoring to one image, you’re giving V8 a collection to pull from. This is useful when you want a blended aesthetic or when no single image captures exactly what you’re after.

Moodboards also default to --sv 7, with the same improvements as srefs: faster, cheaper, and now compatible with --hd, personalization, --stylize, and --exp. Use --sv 6 to revert.

How to get the best results right now

a lone figure encountering a glacier calving into black water, approaching storm, cinematic scale and smallness, wide-to-intimate composition shift with a lone figure tiny against immensity or immensity reflected in their face, shallow depth of field alternating between awe-struck figure and the thing causing the awe, light that matches the sublime of the scene depending on the sublime, very fine film grain adding fragility to the moment, mouth slightly open, body forgetting itself, devotional light quality, color grading that feels like a held breath, prestige contemplative cinema, the ego briefly dissolved by beauty — Generated in Midjourney v8

Midjourney’s own recommendation: use your personalization profile and push --stylize up to 1000. V8 is calibrated to perform best when the stylization systems are doing heavy lifting.

For photorealistic or controlled output: switch to --raw immediately, or anchor the aesthetic with a moodboard or sref. Without it, V8’s default aesthetic can lean expressive.

Prompts: longer and more specific works better in V8. If V7 rewarded brevity, V8 rewards detail.

The new Grid Mode in the interface is worth using; you can generate a large set of thumbnails quickly, browse them in one view, then upscale only the ones that are worth it.

🎥 Comparing Midjourney v7 to v8

🖼️ Midjourney v8 Prompts

Prompt: Photorealistic yakuza drama still, gang members in sharp black suits gathered in a dimly lit izakaya, cigarette smoke curling, sake bottles on table, tense standoff energy, dramatic shadows, desaturated color palette with red accents, film grain, 8k resolution --chaos 10 --ar 16:9 --sref 8416398201 --raw --sw 50 --v 8

Prompt:
Retrowave calico cat with glasses character design,sticker,Full body shot ,anime style, Trigger Studio style, manga art, comics, inking, graffiti art, graphic, neon colors, golden ratio composition, design for tshirt --chaos 10 --ar 16:9 --raw --v 8 --stylize 300

🎨 Explore my Portfolio

I’ve had some amazing opportunities to work on a variety of projects recently, and I don’t share everything in this newsletter or online. And you know I love testing tools to share.

Take a look at my portfolio to get a quick glimpse of my work:

EXPLORE MY PORTFOLIO

🚀 My Recent Top AI Tools Picks

Magnific Video Upscaler: Upscales video up to 4x resolution while enhancing detail and reducing compression artifacts. Currently in early access with a queue system; very impressive results, and it comes with the same "creativity" slider approach as the image upscaler.
Kie.ai: One API to access all the best AI models in one place. Veo 3.1, Runway Aleph, Nano Banana, Suno, Flux Kontext and more, all through a single credit-based system. If you’re building apps or automating creative workflows, this saves you from juggling multiple subscriptions and API keys. Cheaper than FAL.
Krea Nodes - Prompt-to-Workflow: Krea just launched Prompt-to-Workflow inside their Nodes editor. Describe what you want to create in a sentence and it builds the entire multi-step pipeline for you: image generation, upscaling, style transfer, video, all connected and ready to run.
Claude Code for VS Code: Anthropic’s AI coding agent now lives directly inside VS Code. It reads your files, proposes changes with inline diffs, and handles everything from debugging to building full features. I’ve been using it to build tools and automations without switching between apps. 5.2 million installs.

That’s the Midjourney v8 overview as it stands. If you’re testing it this week, start with a detailed, literal prompt—describe your subject, setting, lighting, camera angle, and mood in plain language. Skip the cryptic tokens and style hacks for now; v8 responds better to clear visual direction than shorthand.

Try this template as your starting point:

[Subject doing action], [setting/environment], [time of day], [lighting type], [camera angle/lens], [mood/atmosphere]

Example:

A woman reading a letter at a wooden desk, cluttered study with tall bookshelves, late afternoon, warm golden light through dusty windows, medium shot at eye level, quiet and contemplative

This gives the model enough specificity to work with while keeping your prompt interpretable and predictable. Once you see how v8 handles the basics, then start experimenting.

And remember: this is still Alpha, not the finished product.

Thanks for reading.

I hope you have a creative week!

Heather Cooper

Discussion about this post

Ready for more?