Full parameter reference, setup guides for Personalize/Moodboards/Style Creator, and prompting rules for Midjourney v1-v8.2.
| Category | Introduced / Version | Parameter | Syntax | Accepted Values / Range | Prompt Example | Description & Notes |
|---|---|---|---|---|---|---|
| 1. Essential | v1 – v8.2 | Version | --version, --v |
1, 2, 3, 4, 5, 5.1, 5.2, 6, 6.1, 7, 8, 8.1, 8.2 | metropolis --v 8.2 |
Selects the model version. V8.2 is the default since July 24, 2026; V8.1 was the default from June 10 to July 23, 2026. The V8.0 alpha has been retired. |
| 1. Essential | v1 – v8.2 | Aspect Ratio | --ar, --aspect |
width:height (e.g. 16:9, 9:16, 1:1, 4:3) | majestic waterfall --ar 16:9 |
Adjusts aspect ratio of rendered images. Supports extreme ratios up to 14:1 or 1:14. |
| 1. Essential | v1 – v8.2 | Seed | --seed |
0–4294967295 | neon city --seed 12345 |
Sets the random starting noise grid seed for generating deterministic, reproducible results. |
| 2. Style Controls | v1 – v8.2 | Stylize | --stylize, --s |
0–1000 (default 100); best 100–400 in v8+ | cyberpunk warrior --stylize 250 |
Controls how strongly Midjourney applies its default artistic aesthetic polish. |
| 2. Style Controls | v6+ | Style Reference | --sref |
Image URL, SREF Code, or random | fantasy castle --sref 3623375 |
Applies visual style, color palette, and artistic texture from reference images or codes. |
| 2. Style Controls | v6+ | Style Weight | --sw |
0–1000 (default 100) | cozy cottage --sref URL --sw 250 |
Sets relative influence strength of the --sref style reference image. |
| 2. Style Controls | v6+ | Style Version | --sv |
1–6 (default 6 in v7+); 1–4 in v6 | dragon --sref URL --sv 4 |
Picks the style-reference algorithm. In v7 and later, --sv 6 is the default and --sv 4 is the old V7 model, useful for older sref codes. Sref codes and --sref random only work with --sv 4 and --sv 6. Not compatible with Moodboards. |
| 2. Style Controls | v5.1+ | Style Raw | --style raw, --raw |
raw | street portrait --raw |
Strips away default auto-beautification for candid, realistic, or photographic fidelity. |
| 2. Style Controls | v6.1+ | Personalization | --profile, --p |
User Profile Code, random | serene landscape --p |
Applies custom trained user preferences, moodboard, or personal rating style profile. |
| 2. Style Controls | v7+ | Experimental | --exp |
Toggle | abstract structure --exp |
Enables experimental creative aesthetic rendering features. |
| 3. Character Controls | v7 only | Omni Reference | --oref |
One image URL | retriever astronaut --oref URL --v 7 |
Puts a person, object, vehicle, or creature from one reference image into a V7 image. Costs 2x GPU time and isn't compatible with --q 4 or Draft mode. In V8.1/V8.2, use the Edit model (--edit) instead; it takes up to four references. |
| 3. Character Controls | v6 (legacy) | Character Reference | --cref |
Image URL | photo of a man --cref URL --v 6.1 |
Legacy V6 character reference. Replaced by Omni Reference in V7 and by the Edit model in V8.x. |
| 3. Character Controls | v6+ | Character Weight | --cw |
0–100 (default 100) | photo of woman --cref URL --cw 0 |
Fidelity weight for --cref (V6): 0 matches face only; 100 matches face, hair, and clothing. Midjourney strips --cw automatically when no character reference is attached. |
| 3. Character Controls | v7 only | Omni Weight | --ow |
1–1000 (default 100) | retriever astronaut --oref URL --ow 250 --v 7 |
Controls how much detail from an Omni Reference carries over. Keep it under 400 unless you're also using a very high --stylize. On the web it's the Omni Strength slider. |
| 3. Character Controls | v8.1 / v8.2 | Edit Model | --edit |
Up to 4 image URLs, space-separated | make this silver knight ride the horse of fire --edit URL1 URL2 |
New (Aug 27, 2026). Edit an image with plain-language instructions, combine up to four reference images, change perspective, retexture, or inpaint and outpaint. Replaces Omni Reference, Character Reference, and Retexture in V8.x. Matches the first image's aspect ratio unless you add --ar. Works with Personalization, Moodboards, srefs, --raw, and --hd; not with --tile. On the web, attach images under “Attach to prompt” or click Edit in the lightbox. |
| 3. Character Controls | v1 – v8.2 | Image Weight | --iw |
0–3 (default varies) | URL sports car --iw 2.0 |
Sets relative visual weight of input image prompt URL vs text prompt. |
| 4. Variation & Prompt | v1 – v8.2 | Chaos | --chaos, --c |
0–100 | mystical creature --chaos 50 |
High chaos values yield highly varied, unexpected, and diverse initial grid options. |
| 4. Variation & Prompt | v5.2+ | Weird | --weird, --w |
0–3000 | alien architecture --weird 1000 |
Introduces quirky, bizarre, and unconventional aesthetic concepts into the generation. |
| 4. Variation & Prompt | v1 – v8.2 | Negative Prompt | --no |
Text string | forest --no text, watermark, blur |
Excludes specified objects, styles, elements, or colors from appearing in the output. |
| 4. Variation & Prompt | v1 – v8.2 | Tile | --tile |
Toggle | floral pattern --tile --v 8.2 |
Generates one seamless tile for wallpapers, fabric, and textures. Since Sept 24, 2026, V8.1/V8.2 tiles blend without visible seams. Not compatible with the Edit model or an attached image. Upscaling can break the seamless edge. On the web, click a tile image twice to preview the repeat. |
| 4. Variation & Prompt | v4+ | Repeat | --repeat, --r |
1–40 (varies by tier) | futuristic vehicle --repeat 5 |
Runs the exact same prompt multiple times in a single batch submission. |
| 5. Resolution & GPU | v8+ | High Definition | --hd |
Toggle | portrait --hd --v 8.2 |
Renders 2K (2048px) images natively, with no upscaling. Costs 1.3 GPU minutes per job, versus 0.8 for SD. HD supports aspect ratios up to 4:1. Pan, Zoom Out, and Vary Region downscale HD results to SD. On the web, Rerun as HD re-renders any SD job. |
| 5. Resolution & GPU | v8.1+ | Standard Definition | --sd |
Toggle | concept art --sd --v 8.1 |
Generates standard 1024px images at 0.8 GPU minutes per job. Fast for exploring; rerun favorites in HD. |
| 5. Resolution & GPU | v1 – v8.2 | Quality | --quality, --q |
.25, .5, 1, 4 | complex scene --q 4 |
Rendering effort. --q 4 in v8+ delivers enhanced coherence mode for complex multi-subject prompts. |
| 5. Resolution & GPU | v7 / v8.1+ | Draft Mode | --draft |
Toggle (⚡ button on the web) | neon koi pond --draft --sref random |
V8.1/V8.2: one prompt makes a grid of 24 images at 512×512 for 0.4 GPU minutes; click Vary or Remix on a draft to render it at full SD or HD. Web only, not Discord. V7: 4 images at half the GPU cost; Enhance re-renders at full quality. Permutations aren't supported. |
| 5. Resolution & GPU | Jun 2026+ | Preview Models | --preview |
Toggle | epic landscape --preview |
Corrected: runs your prompt on an early, unreleased model so you can test upcoming features. Results can be unpolished and aren't guaranteed to stay consistent over time. Expect differences with Personalization and Moodboards. |
| 5. Resolution & GPU | v1 – v8.2 | Stop Generation | --stop |
10–100 | abstract portrait --stop 80 |
Stops generation process early at the specified percentage for softer or artistic incomplete renders. |
| 5. Resolution & GPU | v1 – v8.2 | Fast Mode | --fast |
Toggle | robot warrior --fast |
Processes job immediately using Fast GPU subscription allowance. |
| 5. Resolution & GPU | v1 – v8.2 | Relax Mode | --relax |
Toggle | mountain lake --relax |
Processes job in queue without spending Fast GPU time. |
| 5. Resolution & GPU | v5.2+ | Turbo Mode | --turbo |
Toggle | urgent render --turbo |
Processes job at maximum velocity using high-speed GPU tier. |
| 6. Video & Motion | Video model | Motion | --motion |
low (default), high | waterfall flowing --motion high |
Low motion favors still scenes, slow camera moves, and subtle character movement. High motion gives bigger camera and character movement but can look glitchy. |
| 6. Video & Motion | Video model | Video | --video |
Toggle (Discord) | IMAGE_URL floating astronaut --video |
Animates an image into a 5-second video. In Discord, put your image URL first and add --video; on the web, use Animate. Extend up to four times by 4 seconds each, to a 21-second maximum. Videos only accept --motion, --raw, --loop, --end, and --bs. |
| 6. Video & Motion | Video model | Loop | --loop |
Toggle | abstract waves --loop |
Configures animated video output to loop seamlessly from start to end frame. |
| 6. Video & Motion | Video model | End Frame | --end |
Image URL / Frame | sunset --end URL |
Specifies target keyframe image URL for morphing transitions in video animations. |
| 6. Video & Motion | Video model | Batch Size | --bs |
1, 2, 4 (default 4) | cinematic sequence --bs 2 |
Number of videos per prompt. Smaller batches save GPU time: an SD batch of 4 costs 8 minutes, a batch of 1 costs 2. |
| 7. Special Models | v4 – v7 | Niji Engine | --niji |
4, 5, 6, 7 | anime scene --niji 7 |
Model tuned for anime and Eastern illustration styles. Niji 7 (Jan 9, 2026) improved coherence, prompt following, and text rendering, and gained Personalization and Moodboards in February 2026. |
| 7. Special Models | Legacy (v3) | Test Models | --test, --testp |
Toggle | portrait --testp |
Legacy artistic and photorealistic test engines used during early Midjourney v3 development. |
| 8. Legacy Options | Legacy (v1–v3) | Sameseed | --sameseed |
Toggle | pattern --sameseed |
Applies identical initial noise grid across all 4 grid images in legacy engines. |
| 8. Legacy Options | Legacy (v1–v5) | Upscalers | --upbeta, --uplight |
Toggle | retro car --upbeta |
Legacy upscaler algorithms from early Midjourney releases prior to external super-resolution tools. |
| 8. Legacy Options | Legacy (v1–v3) | Pixel Dimensions | --w, --h |
Pixels (e.g. 800, 600) | room --w 800 --h 600 |
Explicit pixel width and height input (Replaced entirely by the modern --ar parameter). |
| 9. Privacy | All versions | Stealth Mode | --stealth |
Toggle | product concept --stealth |
Keeps new creations private on the Midjourney website. Images you make in a shared folder are still public unless they're made in Stealth mode. |
| 9. Privacy | All versions | Public Mode | --public |
Toggle | poster design --public |
Makes creations public on the Midjourney website, the opposite of --stealth. |
Prompt Architecture & Descriptive Vocabulary
1. Primary Subject
Define the core focal point with crisp clarity (e.g., an armored mechanical paladin, an overgrown gothic greenhouse).
2. Context & Setting
Establish placement, action, or physical backdrop (e.g., perched atop a rain-slicked ledge, suspended in zero gravity).
3. Artistic Medium
Designate the foundational art form or genre (e.g., minimalist vector graphics, dark surrealism, 1990s anime film screen).
4. Modifiers & Atmosphere
Layer nuance using tailored lighting, optics, or textures (e.g., volumetric haze, bioluminescent reflections).
5. System Parameters
Append execution directives at the end (e.g., --ar 16:9 --stylize 250 --v 8.2).
Illumination & Lighting (A–Z)
Color Dynamics & Palettes (A–Z)
Optics, Framing & Angles (A–Z)
Surfaces & Material Textures (A–Z)
Environmental Mood & Atmosphere (A–Z)
Art Movements & Genres (A–Z)
Top 20 Signature Artist Influences (A–Z)
Cool Tips & Tricks
--s {50, 250, 500, 1000}
Note: Permutations cannot be used in draft mode.
A kid-friendly educational drawing video on a clean white sheet of paper. A black marker slowly draws a beautiful {rose, orchid, sunflower, tulip} flower
Note: Permutations cannot be used in draft mode.
--sref random --draft
hand-drawn botanical pattern --tile --v 8.2
Note: --tile doesn't work with the Edit model or an attached image.
Prompt Engineering Framework & Rules
To guarantee full-body coherence and prevent missing limbs, cropped heads, or anatomical glitches in Midjourney, ALL prompt outputs must strictly incorporate these structural anchors:
- Explicit Full-Body Framing: Always start with
"A full-length sticker art fashion illustration..."and include"Full body shot"near the end. - Defined Limb Placement: Direct the limbs explicitly (e.g.,
"arms extended"or"arms extended in a dynamic balance pose"). Never leave hand or arm positioning unguided. - Solid Background Anchor: Place the subject on a
"solid plain background"or"simple [color] background". Never put busy background elements directly behind the character; use them only as color/vibe inspiration. - Sticker Outline Guarantee: End the prompt text with
"completely surrounded by a thick white border".
Unless specified otherwise by the user, default to the following aesthetic rules:
- Style Engine: 1990s editorial magazine aesthetic, minimalist vector style, clean ink lines, flat design, sophisticated color blocking.
- Regional Theme: San Pedro, CA & Coastal California. Derive color blocking and outfits from maritime, industrial port, ocean-sunset, and retro coastal streetwear vibes (e.g., Cabrillo Beach, Korean Friendship Bell, Point Fermin, Port of LA).
Appended to the end of every single prompt without exception:
- Aspect Ratio:
--ar 77:58(or user-defined ratio such as1:1). - Stylization Weight:
--stylize 100(prevents wild over-stylization and preserves clean vector geometry).
What's New in Midjourney (2026 Highlights)
Thinking Mode (Alpha, Oct 8, 2026)
Rerun a V8.2 job (standard or edit) “with thinking”: Midjourney looks at the image, finds where it missed your prompt (wrong objects, layout, anatomy, or text), and generates again. Open an image on alpha.midjourney.com and click Rerun (Thinking) in the lower right.
Shared Folders (Alpha, Oct 7, 2026)
Invite people by username as Collaborators, who can generate straight into the folder, or Viewers, who can browse. You can also share by link. Images stay in each creator's own history.
- Privacy: a private folder doesn't make images private. Non-stealth images can still appear on Explore and profiles.
Live Style Previews & Default Parameters (Sept 2026)
Turn on Live previews in the Styles sidebar to see your current prompt rendered across liked and featured styles, then click one to generate with it. Save your usual settings under Settings → Advanced → Your defaults, or click “Set current pills as default”. Folders can keep their own default parameters, including --exp.
Edit Model for V8 (Aug 27, 2026)
Edit images with plain-language instructions, combine up to four reference images, and inpaint or outpaint. It replaces Omni Reference, Character Reference, and Retexture in V8.x. Since Sept 24, edits change only the pixels you select, so repeated edits don't degrade the image.
V8.2 Becomes the Default (Jul 24, 2026)
Focused on aesthetics, image quality, and Personalization: bolder, more creative images, far fewer random low-quality results, and profiles that understand your taste much better.
Draft Mode for V8.1+ & --preview (Jun 16, 2026)
Draft prompts make 24 low-resolution images; Vary the ones you like for a full render. --preview lets you test early versions of upcoming models. Add --sref random to a draft to get 24 different styles (Jun 25).
V8.1: Faster, Cheaper HD (Apr–Jun 2026)
V8.1 became the default on June 10. SD images appear in about 4 seconds and HD in about 12; HD renders natively at 2K. It also brought back image prompts and image weights, a Prompt Shortener, and a more detailed Describe.
Korean, Niji 7 & More
Korean is live on midjourney.com (Sept 2026). Niji 7 launched January 9, 2026 and gained Personalization and Moodboards in February. Four aspect ratios were added to the slider in January: 6:11, 4:5, 5:4, and 21:9.
Beginner Guide: How to Setup Personalize (--p / --profile)
Rank Images to Train Your Model
Go to midjourney.com/personalize. Since February 2026 you build a profile by clicking the images you like as you scroll a list, instead of choosing between two images at a time. V8.2 profiles draw from a much larger pool of images.
- Minimum threshold: Rank at least 100 to 200 pairs to initialize your personal model.
- Tip: The more images you rank (e.g., 500+), the more nuanced and accurate your personalized profile becomes.
Apply Personalization in Your Prompts
Once you've ranked enough images, simply append --p or --profile to any prompt in Discord or the Web Interface:
Midjourney will automatically inject your unique artistic DNA into the render!
Obtain & Share Your Unique Profile Code
Midjourney assigns an alphanumeric code to your trained profile (e.g., --p 7a2b9x). You can copy your code from your account settings to share with friends or lock in a specific iteration of your aesthetic.
Adjust Personalization Strength with Stylize (--s)
You can dial up or dial down how strongly your profile influences the output using the Stylize parameter:
--s 50(Subtle): Very light touch of your profile.--s 100(Default): Balanced mix of prompt + profile.--s 500–--s 1000(Heavy): Strong application of your personal aesthetic.
Enable Personalization by Default
If you love your personalized look and want it on every render, type /settings in Discord or open settings on the Web App and toggle Personalization ON.
Beginner Guide: How to Setup & Use Moodboards
Open Moodboards in the Web Interface
Log in to Midjourney.com, go to your Organize or Styles / Moodboards tab in the sidebar navigation, and click the "+ Create New Moodboard" button.
Curate & Upload Reference Images
Drag and drop 5 to 15 high-quality images into your new moodboard. Name your board something descriptive like 70s Retro Sci-Fi, Pastel Architecture, or Neon Cyberpunk.
- Include color swatches for palette guidance.
- Add lighting or material texture close-ups (e.g. brushed metal, satin drapery).
Generate or Copy Your Moodboard Style Reference Code
Once your moodboard is populated, click "Use as Style Reference" or copy the generated --sref URL / code linked to your moodboard collection.
Control Moodboard Influence with Style Weight (--sw)
Adjust how aggressively Midjourney adopts your moodboard aesthetics:
--sw 50: Subtle color and atmosphere inspiration.--sw 100(Default): Standard style transfer.--sw 300+: Overwhelming aesthetic takeover matching the moodboard texture.
Beginner Guide: How to Master the Style Creator
Launch the Style Creator Workspace
On Midjourney.com, open the Styles Manager or Style Creator section. Click "Create New Style Preset".
Blend Multiple References & Set Weights
Select 2 to 4 style reference images or enter existing --sref codes. You can adjust individual strength multipliers for each component image:
- Image A (Line art): Weight 1.0
- Image B (Pastel colors): Weight 0.5
- Style Version: Choose
--sv 4for modern high-fidelity rendering.
Test Your Preset in the Sandbox
Use the built-in Sandbox Test Prompt field (e.g., a cat sitting on a bench or a sports car on a highway) to preview how your blended style looks across diverse subjects before saving.
Save & Export Your Custom Style Code
Give your style preset a memorable title (e.g., Muted Editorial Vector). Click Save to generate your custom reusable --sref code or alias: