Midjourney V8 Prompting: What Changed from V7

Updated April 15, 2026: V8.1 Alpha shipped April 14.¹¹ The big V8.1 changes that affect the V8 prompting playbook:

• HD mode is now default (3x faster and 3x cheaper than V8.0’s HD): no separate flag, no 4x surcharge
• Standard resolution: 50% faster and 25% cheaper than V8.0 SD; full quality now matches V7 draft speed
• “Run as HD” button reruns the seed-locked SD prompt in HD
• Image prompts and image weights are back (V8.0 broke them)
• Prompt Shortener auto-engages when a prompt exceeds the length limit
• Describe rewritten to produce longer, more detailed V8-style prompts
• Moodboards and srefs are “super stable” now
• Aesthetic pulled back toward V7, less of the hyper-polished V8.0 look
• V8.0 is likely to be decommissioned “after V8.1 has been out for a few weeks” (Midjourney’s phrasing, not a committed date)¹¹

The thesis below (prompt → profile shift) still holds. The economics arguments from the original March 18 post have been updated inline. Section-by-section notes: The Cost Math Changed, HD as Default Output, Longer Prompts Actually Work, and a new Image Prompts and Image Weights Return section.

Midjourney V8 Alpha launched on March 17, 2026, and V8.1 Alpha replaced V8.0 as the current V8-series model on April 14, 2026.¹¹¹ The immediate reactions focused on speed (5x faster) and resolution (native 2K). Those are real improvements. They are also not the point. V8 changes the creative loop itself: how you explore, what you invest in, and where your prompt even matters relative to the system’s own understanding of your taste. If you are prompting V8 the same way you prompted V7, you are working against the tool.

Midjourney V8 shifts creative control from your text prompt to your personalization profile. The model performs best at --stylize 1000 with a trained profile. V8.1’s standard resolution is so fast it matches V7 draft speed at full quality,¹¹ and HD (native 2K) is now the default output with a 3x cost reduction from V8.0.¹¹ The workflow: explore at standard resolution, let “Run as HD” upgrade any winning grid, and let your profile handle aesthetic direction.

I maintain a full Midjourney guide covering V7, Niji 7, video generation, and the complete parameter set. The post you are reading focuses on what V8 changes and why your V7 habits need to adapt.

TL;DR

Stop prompting V8 like V7. V8 shifts the creative center of gravity from your prompt text to your personalization profile, so aim for --stylize 1000 with a trained profile, per Midjourney’s own recommendation.¹ Explore at standard resolution, use “Run as HD” to rerun the seed-locked prompt in HD when a grid is worth keeping,¹¹ and let conversation mode handle prompt construction when you’d rather talk through ideas than write them.³ Skip --q 4: it is a V7-era parameter that V8.1 does not support.¹³ For character consistency, reach for Personalization and srefs instead of --cref/--oref: V8.1 supports neither.¹³

The Exploration Loop Changed

V8’s architecture supports a different creative loop from the traditional 4-image grid: generate many images at standard resolution quickly, browse them in the new Grid Mode, then rerun winners at HD. The interface now includes sidebar settings that stay visible without blocking your workspace, plus reference thumbnails and Personalization profiles anchored alongside the Imagine bar.¹¹³

The shift matters for two reasons. First, quantity changes how you explore. Four images force you to evaluate each one. Generating rapidly at standard resolution lets you scan for the version that matches the image in your head, the one you could not have described precisely enough to prompt for. Second, the economics are mild. V8.1 made standard resolution 50% faster and 25% cheaper than V8.0, fast enough at full quality to match V7 draft speed.¹¹ HD is default, so there is no 4x surcharge for sharp finals the way there was in V8.0.¹¹ The new “Run as HD” button reruns the seed-locked SD prompt in HD, which makes the explore-then-commit loop concrete.¹¹¹³

The V7 workflow I recommended in the Midjourney guide — 60% Draft, 30% Fast, 10% Final — does not map onto V8. Draft mode is less necessary when V8.1’s standard resolution is already as fast as V7 draft speed at full quality. The new allocation: explore freely at standard, then “Run as HD” on the grids worth keeping.

Personalization Replaced Your Prompt

V8’s most significant change is not a parameter. The model now leans on your personalization profile (the aesthetic preferences Midjourney learned from your image ratings) more than on your text prompt. Midjourney’s official guidance: “lean heavily into personalization” and “crank --stylize up to 1000.”¹

In V7, your prompt was the primary creative input. You described what you wanted, and the model interpreted it through a default aesthetic. In V8, your profile is the primary creative input. The prompt provides subject and context. The profile provides aesthetic direction. A well-trained profile at --stylize 1000 produces more coherent results than a perfectly written prompt at --stylize 100.

If you have an existing V7 personalization profile, V8 carries it over automatically with a few additional ratings needed for V8 compatibility.⁵ Midjourney’s personalization system activates at 40 ratings and stabilizes around 200 ratings, with continued improvement up to 2,000.⁶ If you have never rated images for your profile, V8 is running at a fraction of its capability. I recommend completing at least 200 ratings before forming opinions about V8 output quality.

The implication for prompt engineering: V8 prompts can focus on what is in the scene rather than how it looks. Aesthetic direction (mood, color temperature, film grain, lighting style) comes from the profile and --stylize value. Scene description (subject, setting, composition) comes from the prompt. Splitting these concerns produces better results than cramming everything into the prompt text.

Conversation Mode: Let the AI Write Your Prompt

V8 introduces a conversational interface where you describe your ideas in natural language (text or voice) and an AI interprets them into the actual prompts used for generation.³ You are not writing prompts directly in this mode. You are directing a creative conversation.

The interaction supports iterative refinement: refer to specific outputs by saying “run a variation of image 4” or “make the lighting warmer in image 2.”³ Conversation mode works in multiple languages. The AI handles the translation from your description to Midjourney’s prompt syntax, including parameter selection.

The practical implication: conversation mode lowers the barrier for people who find prompt engineering intimidating. For experienced users, the value is different: the AI sometimes interprets your intent in ways you would not have prompted for directly, surfacing unexpected directions. Consider it a creative collaborator rather than a shortcut.

For precise control, bypass conversation mode and write prompts directly in the Imagine bar. The two approaches complement each other: conversation mode for exploration, direct prompts for execution. The pattern mirrors a broader shift in creative tools where taste becomes the primary input rather than technical syntax.

Longer Prompts Actually Work

V7 rewarded concise, natural language descriptions. Keyword stuffing degraded results. V8 reverses this partially. The model follows detailed directions more reliably and holds onto small details that V7 would drop.¹

Midjourney’s own recommendation for V8: “trend towards longer, more specific prompting.”¹ The improved language understanding means complex multi-subject scenes with spatial relationships produce more coherent results than in V7. Complex negative parameter chains (--no sky --no clouds --no birds) can often be replaced with plain English: “an interior scene with no visible windows.”

V8 still processes tokens sequentially with front-loaded weighting, so words at the beginning of the prompt carry more influence.⁷ The difference is that V8 processes the full prompt with better comprehension. A V7 prompt that worked at 15 words might work better at 40 words in V8 if those additional words add meaningful scene direction.

A practical V8 prompt structure:

[Subject and action], [environment and context], [specific details V8 should preserve],
[lighting and atmosphere] --s 1000 --p --ar 16:9

V8.1 added a Prompt Shortener that auto-engages when a prompt exceeds the length limit.¹¹ Practically: you can draft longer descriptions without guessing at the cutoff, and the system will condense overruns rather than silently truncate. If you want full control over what survives the cut, keep prompts inside the limit yourself. The Shortener is a safety net, not a replacement for editing.

Raw Mode in V8.1

V8.0 shipped with an aesthetic many users called over-polished and hyper-stylized. Midjourney itself acknowledged the “standard V8 aesthetic isn’t finished yet.”¹ V8.1 is Midjourney’s direct answer: Midjourney tuned the V8.1 default aesthetic to feel “consistent and familiar in the spirit of V7.”¹¹ Moodboards and srefs are also “super stable” now, so style anchoring works reliably.¹¹

The V8.1 release changes the raw-mode recommendation. In V8.0, --raw was an escape hatch from a drifting default. In V8.1, you reach for --raw when you want to strip styling, not to avoid instability. For photorealistic work, architectural visualization, or editorial illustration where the default still feels too produced, --raw removes the model’s interpretive layer and takes the prompt more literally.

The combination of --raw with high --stylize restores structure while preserving expressive qualities: form and layout hold, but textures and details get the V8 treatment.⁵ For exploratory creative work where you want V8’s aesthetic, leave --raw off and rely on the V8.1 default + your personalization profile.

Style Creator: Build Your Visual DNA

The Style Creator generates custom --sref codes through visual preference selection: pick images from grids that match your desired aesthetic, and the system creates a reusable style code without words.⁸ Each session runs through refinement rounds: early rounds vary widely as the system narrows in on your taste, stabilizing around rounds 5-10, with subtle refinements through round 15.⁸

A practical tip: start the Style Creator with the simplest possible prompt, even just a period (.), to produce a style code that transfers flexibly across different prompts.⁸ Complex starting prompts lock the style code to that specific subject matter, reducing transferability.

Generated style codes function as custom --sref values you can reuse across any prompt. Multiple codes can be stacked (e.g., --sref 7031655206 4802911573) to blend styles.⁸ The Style Creator’s preview images currently use V7 for generation; V8 Style Creator support has not shipped.

Style Creator uses GPU time: the preview images generated during each round count against your allocation. Budget accordingly when creating multiple custom styles.

Text Rendering Works

V7 text rendering was a coin flip. V8 renders text reliably when you follow two rules: wrap the text in double quotes within the prompt, and keep the text short.⁹

A prompt like a neon sign saying "OPEN 24/7" on a rainy street at night produces clean, readable text in V8. Single quotes do not work; only double quotes trigger the text rendering pipeline. Adding context words (“sign saying,” “poster reading,” “text written”) improves placement reliability.

Longer text strings are possible in V8 but less reliable. Two to four words render consistently. Full sentences degrade. For text-heavy applications (mockup signage, title cards, poster designs), V8 is usable for headlines but not body copy. Ideogram V2 remains more accurate for precise text rendering, while V8 excels at artistic context around text rather than typographic precision.¹⁰

If text gets stylized beyond readability, lower --stylize or switch to --raw. High stylization values can treat text as a visual element rather than readable content.⁹

HD as Default Output

Previous Midjourney versions generated at 1024x1024 and upscaled to higher resolutions, introducing artifacts and softening details along the way. V8 generates natively at 2048x2048.² Midjourney trained the model for high-resolution output, and the details at 2K are genuinely sharper, not interpolated.

V8.0 treated HD as a premium tier: --hd jobs cost 4x more and ran 4x slower than standard.⁴ V8.1 changed this. HD is 3x faster and 3x cheaper than it was in V8.0, which Midjourney says is “so cheap we’re making it default for V8.1.”¹¹ HD in V8.1 costs about 1.33 minutes of GPU time; SD costs less than 1 minute.¹³ HD is still the more expensive tier, just not dramatically so, and it ships as the default rather than an opt-in flag.

The practical workflow shift: V8.0 forced you to choose resolution up front based on cost. V8.1 defaults to 2K output and lets SD stay as the cheaper fast-iteration mode. You iterate in SD, use the “Run as HD” button to rerun the seed-locked prompt in HD when a grid is worth keeping, and skip a separate upscaling workflow entirely.¹¹¹³

With srefs and moodboards now on the “super stable” --sv 7 path, the old V8.0 constraint that HD couldn’t combine with style references no longer shapes the daily workflow.¹¹¹²

The Cost Math Changed

V8.0 restructured the economics of exploration versus refinement by making HD a 4x premium. V8.1 pushed the shift further: HD is default and dramatically cheaper than in V8.0, and SD itself got faster and cheaper.

Relax mode is available for Standard, Pro, and Mega subscribers in V8.1.¹³ The March 21 V8.0 exception (no Relax for --hd + --q 4) no longer applies because V8.1 does not support --q 4 at all.¹³

The practical impact: SD at full quality in V8.1 is fast enough (under a minute of GPU time) that “exploration” and “refinement” stop being separate workflow phases. HD costs roughly 1.3x an SD job, not 4x, so final output is cheap enough to reach for often. The “Run as HD” button lets you reach HD without re-prompting; it reruns the seed-locked SD prompt in HD.¹¹¹³

Image Prompts and Image Weights Return

V8.0 shipped without working image prompts or image weights, two workflows that existed in V6 and V7 and that a large part of the community depends on. V8.1 brought them back.¹¹ Midjourney frames the return as part of “some community favorite features” restored in the release.¹¹

Why this matters for V8 prompting: image prompts let you seed generation with a visual reference beyond text, and image weights (--iw) let you control how much that reference influences the output relative to the prompt. For reference-driven workflows (style-matching an existing photo, continuing a sequence, incorporating a specific texture or composition), these are foundational primitives. If you built a workflow around image prompts and stopped using V8 when V8.0 broke them, V8.1 is the release that reopens that door.

The restoration pairs with --sv 7 moodboards and srefs being “super stable” in V8.1.¹¹ The three reference systems (image prompts, srefs, moodboards) are now coherent in V8 for the first time.

Character Consistency: What Works in V8 (and What Doesn’t)

V8.1 Alpha does not support --cref or --oref per Midjourney’s compatibility chart.¹³ That gap matters because character consistency is a real workflow need and V8 offers no direct parameter for it today.

Here’s the state of play by version:

• V6 / Niji 6: --cref (Character Reference) with --cw (Character Weight)
• V7: --oref (Omni Reference) with --ow (Omni Reference Weight), introduced in V7 to replace --cref
• V8.1 Alpha: neither works today¹³
• Niji 7: no character/omni reference; Midjourney has hinted at a replacement

What to use for character consistency in V8.1 while omni reference is absent:

• Personalization profile: a well-trained profile provides aesthetic continuity across generations, which carries some character-level stability when you describe the character consistently
• Image prompts (restored in V8.1):¹¹ seed with a reference image and let the model carry visual features forward
• Srefs and moodboards on --sv 7: anchor style (lighting, palette, linework) which indirectly stabilizes how a character renders¹²
• Seed control: use --seed to lock visual variance across a series; V8.1 seeds produce 99% identical results given the same prompt¹³

For projects requiring strict character consistency across many images (storyboards, brand mascots, sequential illustration), the honest answer today is: do character-critical work in V7 with --oref + --ow, or wait for V8’s character reference path to land. The original Midjourney announcement for V8.1 flagged upscalers, edit, inpaint, and outpaint as next. Character reference is not on that short list, so do not plan around it shipping soon.¹¹

Where V8 Fits Among Image Generators

V8’s strength is artistic interpretation: atmospheric lighting, emotional resonance, and material detail handled with a stylistic point of view rather than a purely literal translation of the prompt. Where other tools optimize for precision, V8 optimizes for aesthetic output. Whether AI tools displace or empower creative professionals is a separate question; see The Design Career That Survives AI.

For workflows where accuracy trumps aesthetics (text-heavy layouts, precise product photography, strict brand compliance), tools like Ideogram (text rendering) or open-source Stable Diffusion variants (fine-tuning control) may produce more reliable results. The practical approach: use V8 for concept work and artistic direction, then switch to specialized tools when precision is the priority.

What Still Does Not Work (as of V8.1, April 2026)

V8.1 resolved the biggest V8.0 gaps (image prompts/weights, aesthetic drift, HD as a premium tier, moodboard/sref instability, Relax availability). What remains open:

• V8 series is still web-only: alpha.midjourney.com, not Discord, not the main site¹¹
• V8 Alpha creations do not appear on the main Midjourney website¹
• V8.0 will likely be decommissioned a few weeks after V8.1. Midjourney asked users to flag any V8.0 output that beats V8.1 in #ideas-and-features before the switch-off¹¹
• Niji V8 not announced: anime/manga workflows remain on Niji 7 (no --cref, Personalization and Moodboards supported)
• Style Creator generates using V7: V8 style creation has not shipped yet
• Upscalers, edit, inpaint, outpaint: Midjourney flagged these as the next V8-series work post-V8.1¹¹

Key Takeaways

If you are coming from V7: Complete your personalization profile (200+ ratings for stability), set --stylize 1000, and write longer, more specific prompts. Midjourney tuned V8.1’s aesthetic to feel V7-like by default, so you no longer need --raw as an aesthetic escape hatch. Reach for it when you want stripped-down literal output.

If you are budget-conscious: V8.1 SD is under 1 minute of GPU time per job; HD is about 1.33 minutes.¹³ The old “every --hd burns 4x your allocation” math is gone. Iterate freely at SD, reach for HD on keepers.

If you work with text: V8 text rendering is genuinely usable for short strings in double quotes. Test with --raw and lower --stylize if text gets overstylized. For high text accuracy, consider Ideogram V2 instead.

If you want creative direction without prompt syntax: Conversation mode lets you describe ideas in natural language. The AI handles prompt construction. Useful for exploration and non-technical collaborators.

If you stopped using V8 when V8.0 broke image prompts: V8.1 brought image prompts and image weights back. Workflows built around visual references work again.¹¹

For the full parameter reference covering V7, Niji 7, personalization, and the complete prompt engineering framework, see the Midjourney guide. For more on how design principles apply to AI-generated work, see the Design Philosophy series and Beauty and Brutalism.

FAQ

Is V8 available in Discord? No. V8 (both V8.0 and V8.1) is only accessible at alpha.midjourney.com. Discord and the main Midjourney site remain on V7.¹¹¹

What’s the difference between V8.0 and V8.1? V8.1 is a point release that kept V8.0’s speed and instruction-following gains but tuned the aesthetic back toward V7, made HD the default output (3x cheaper and 3x faster than V8.0 HD), made standard resolution 50% faster and 25% cheaper, restored image prompts and image weights, stabilized moodboards and srefs, and added a Prompt Shortener and updated Describe.¹¹ Midjourney says V8.0 is likely to be decommissioned a few weeks after V8.1.

Do my V7 personalization profiles work in V8? Yes. V7 profiles carry over automatically. Complete a few additional image ratings to enable V8 compatibility.⁵

How many ratings do I need for effective personalization? Midjourney’s system activates at 40 ratings and stabilizes around 200. Performance continues improving up to 2,000 ratings, but returns diminish past that point.⁶

Should I rewrite all my V7 prompts for V8? Not necessarily. V7 prompts work in V8, but they leave performance on the table. The biggest improvement comes from adding --s 1000 --p to use personalization, not from rewriting the text.

Do I still need to add --hd for native 2K in V8.1? No. HD is the default output in V8.1.¹¹ Standard resolution is now a fast-iteration mode below the default. The “Run as HD” button reruns the seed-locked SD prompt in HD after the fact.¹³

Is Relax mode available in V8? Yes. Relax mode is available for Standard, Pro, and Mega subscribers in V8.1 Alpha.¹³ V8.0 had a “no Relax on --hd + --q 4 combined” exception; since V8.1 doesn’t support --q 4 at all, that exception no longer applies.

Will the V8 aesthetic keep changing? Less than it did during V8.0. Midjourney tuned V8.1’s aesthetic to feel “consistent and familiar in the spirit of V7,” a direct response to V8.0 aesthetic drift complaints.¹¹ The page still notes V8-series is in “an early testing phase,” so things can change.

How do I get character consistency in V8.1? V8.1 Alpha does not support --cref or --oref per Midjourney’s compatibility chart.¹³ For V8.1 workflows, use Personalization, image prompts (restored in V8.1), srefs/moodboards on --sv 7, and --seed for deterministic variants. For strict character consistency across many images, do the character-critical work in V7 with --oref + --ow until V8’s character reference path lands. --cref is a V6/Niji 6 parameter.

How does Style Creator differ from --sref with random codes? Style Creator builds codes through deliberate visual preference selection (5-15 rounds), while random --sref codes are unpredictable. Creator codes are more intentional but still function as standard --sref values.⁸ In V8.1 the underlying --sv 7 sref system is “super stable,” which makes Style Creator codes more reproducible across prompts.¹¹

Sources