exploreKeep prompts short and visual
Phrases that clearly describe what you want to see work better than long instruction lists or abstract mood words alone.
Example
Foggy redwood trail, sun rays through trees, 35mm film look, mist on ferns --ar 3:2
Midjourney
AI image and video generation from text prompts and reference images.
ImageVideoDesign
How it works / How to use
AI image and video generation from text prompts and reference images. On midjourney.com, type a prompt in the Imagine bar on the Create page to generate a grid of four images. Refine with variations, the Editor, and parameters; video and plan limits depend on your subscription.
- Sign in at midjourney.com and open the Create page.
- Type your prompt in the Imagine bar—or use /imagine on Discord if you prefer the bot there.
- Press Enter (or the send button on mobile) and wait for the four-image grid to finish.
- Select an image to upscale, vary, remix in the Editor, or refine with a follow-up prompt.
Text-to-image generation
Use short, clear phrases that describe one scene. Put parameters such as aspect ratio at the very end after a space.
What to provide
- Subject, setting, and mood you want to see
- Style cues when they matter (medium, lighting, camera feel)
- Anything to exclude from the scene
Details that improve the result
- Short and simple prompts typically generate the best images
- Each generation returns four images in one grid
- Fewer details mean more variety; include specifics when control matters
Example prompt
A quiet street at dusk, wet cobblestones, warm shop lights, shallow depth of field, cinematic photography --ar 16:9
If the first output is not good
- “Same scene, but morning light and empty street.”
- Use Vary (Subtle) or Vary (Strong) on a chosen image
- Add --no rain if unwanted elements keep appearing
Common mistakes
- Writing long lists of unrelated ideas in one prompt
- Placing parameters in the middle of the prompt text
- Expecting exact text or logos without reference images
Refining with variations and upscales
Pick one image from the grid, then use variation or upscale controls on the website—or corresponding actions via Discord—to iterate from that version.
What to provide
- The image grid or single image you want to build on
- Whether you want a small tweak or a larger change
- Elements that must stay the same
Details that improve the result
- Iterate from the closest good result instead of rewriting from scratch
- One major change per pass is easier to control than many at once
- Fast GPU time is included in each plan; check your plan for limits
Example prompt
Start from the top-left image in your grid. Apply a subtle variation, then upscale the version with the best composition.
If the first output is not good
- “Keep the architecture; change only the sky to overcast.”
- Run another subtle vary if hands or faces look wrong
- Remix in the Editor when you need localized edits
Common mistakes
- Abandoning a strong base image after one imperfect variation
- Stacking many unrelated changes before reviewing the first result
- Forgetting which seed or version you liked best
Image prompts and references
Add images through the image icon in the Imagine bar, or include image URLs in Discord prompts. Combine with text that says what the reference should influence.
What to provide
- Reference image(s) for style, subject, or composition
- Text describing what to keep vs what to change
- Whether the reference is for style, content, or likeness
Details that improve the result
- Image prompts influence style and content of new creations
- Style References and Omni Reference (--oref) guide look or subject likeness
- Text still matters—say what should change relative to the reference
Example prompt
Upload a photo of a ceramic mug. Prompt: product photo of this mug on a wooden desk, soft window light, minimal background, commercial photography --ar 4:5
If the first output is not good
- “Same mug and angle; change background to solid sage green.”
- Lower reference weight if the output copies too much
- Add --no text if labels or watermarks appear
Common mistakes
- Uploading a reference but giving no text direction
- Mixing several unrelated reference images without saying their roles
- Expecting readable brand text from references alone
Parameters and aspect ratio
Write the scene first, then add parameters at the end with correct spacing: a space before dashes, no punctuation inside parameters.
What to provide
- Desired aspect ratio or orientation
- Optional controls: chaos, quality, personalization profile
- Elements to exclude with --no
Details that improve the result
- Use --aspect or --ar to change shape (images start as squares by default)
- Use --chaos or --c to increase variety between results
- Use --profile or --p for a personalization profile when you have one
Example prompt
Vibrant California poppies in a field, golden hour, macro photography --ar 2:3 --chaos 20 --no people
If the first output is not good
- “Keep the subject; try --ar 1:1 for social posts.”
- “Reduce chaos for more consistent compositions.”
- Add --quality only when you need higher detail and accept longer processing
Common mistakes
- vibrant California poppies--ar 2:3 (missing space before dashes)
- Putting prompt words after parameters
- Using commas or periods inside parameter blocks
Multi-prompts and weights
Separate ideas with :: (no space before ::, one space after). Add numeric weights after :: when one idea should dominate.
What to provide
- Two or more concepts to blend
- Relative importance of each concept
- Any parameters that apply to the whole prompt
Details that improve the result
- space:: ship treats “space” and “ship” as separate ideas
- space::2 ship makes “space” twice as important as “ship”
- Parameters still go at the very end of the full prompt
Example prompt
forest::2 cyberpunk city::1 neon signs, fog, wide angle --ar 21:9
If the first output is not good
- Hover the prompt after generation to inspect multi-prompt weights
- Adjust weights up or down on the weaker or stronger element
- Simplify to a single phrase if the blend looks muddled
Common mistakes
- Using :: with spaces on the wrong side
- Adding too many weighted segments for one image
- Expecting precise typography from multi-prompt blends
Editor: remix, vary, and edit
Open the Midjourney Editor on the website to remix, vary, or edit generated or uploaded images instead of describing the whole scene again from scratch.
What to provide
- A generated or uploaded image to edit
- The specific change or region to adjust
- What must remain unchanged
Details that improve the result
- Describe edits as deltas, not full scene rewrites
- One focused edit per pass usually works better than many
- Name locked elements: face, pose, palette, background
Example prompt
In the Editor, select the generated portrait and remix: change jacket color to navy, keep face and background unchanged.
If the first output is not good
- “Fix only the hand artifact; keep everything else.”
- “Make the edit more subtle.”
- Return to the closest good version before trying a bigger change
Common mistakes
- Asking for a completely different scene while calling it an edit
- Stacking unrelated edits before checking the first result
- Uploading low-resolution sources and expecting fine detail
Video generation
Use Midjourney’s video tools from the website when your plan includes video generation. Start from a clear visual concept and iterate from the best frame or clip.
What to provide
- Starting image or video concept
- Motion or action you want in the clip
- Plan eligibility for video and Relax modes
Details that improve the result
- Official plans include video generation with plan-specific GPU time
- Pro and Mega plans include unlimited Relax SD video per official plan comparison
- Treat video like image work: iterate from close results
Example prompt
From your best still of a lighthouse at dawn, generate a short clip with slow panning motion and consistent lighting.
If the first output is not good
- “Reduce motion speed; keep camera stable.”
- Regenerate from a sharper still if motion smears detail
- Check remaining Fast GPU time before long video batches
Common mistakes
- Assuming video is unlimited on every plan tier
- Combining many camera moves in one generation
- Starting video from a still with ambiguous subject boundaries
Personalization profiles
Add --profile or --p with your profile name at the end of the prompt after the scene description.
What to provide
- A personalization profile or moodboard you have trained
- Scene text that should adopt your custom style
- Fallback plan if the profile is too strong or weak
Details that improve the result
- Personalization creates custom image styles tied to your account
- Combine profiles with clear subject text for controlled results
- Reduce other style adjectives if the profile already dominates
Example prompt
Editorial fashion portrait, soft studio light, minimal backdrop --profile MyBrand --ar 2:3
If the first output is not good
- “Same profile; simpler background.”
- Remove conflicting style words if output looks overstyled
- Try without the profile to compare baseline Midjourney style
Common mistakes
- Using a profile name you have not created
- Fighting the profile with many contradictory style terms
- Expecting exact brand guidelines without reference assets
How to prompt
Midjourney turns short, visual prompts into images. Describe one scene with concrete details, put parameters last, and iterate from the four-image grid with variations or the Editor.
Subject: ceramic coffee mug on a wooden desk. Light: soft morning window light. Style: product photography, minimal. Exclude: text, logos, hands. Parameters: --ar 4:5 --no watermark
Put parameters at the end
Aspect ratio, chaos, quality, personalization, and --no belong after the scene text, with a space before the dashes and no punctuation inside parameters.
Example
Desert highway at noon, heat shimmer, wide shot --ar 21:9 --chaos 15 --no cars
Separate blended concepts with ::
When mixing ideas, use multi-prompts so Midjourney weighs each part separately before combining them.
Example
watercolor::2 botanical illustration::1 white background --ar 1:1
Use references intentionally
Upload or link reference images when style or subject matter should carry over, and say in text what the reference controls.
Example
[upload room photo] Redesign this room in Scandinavian style, keep window placement, brighter daylight --ar 16:9
Iterate instead of restarting
Pick the best of four, vary or upscale, then remix or edit. Small follow-ups beat rewriting the entire prompt when you are close.
Example
From image 3: Vary Subtle, then remix to change sky from sunset to overcast only.
Best output tips
Start on the Create page
The Imagine bar on midjourney.com is the primary web workflow: type a prompt, generate four images in real time, then refine from the grid.
Prefer short, clear phrases
Midjourney’s docs recommend short, simple prompts. Think snapshot descriptions: subject, setting, lighting, and medium—not paragraphs of instructions.
Format parameters correctly
Put --ar, --chaos, --quality, --profile, --no, and --oref after the text with a space before the dashes. Avoid punctuation inside parameter blocks.
Choose aspect ratio deliberately
Images default to square. Add --aspect or --ar early in planning when the output target (story, banner, print) is fixed.
Control variety with chaos
Higher --chaos values produce more unexpected grids. Lower chaos when you want repeatable compositions for a series.
Exclude with --no
When the same unwanted object keeps appearing, --no text, --no watermark, or --no people is clearer than repeating exclusions in prose.
Use multi-prompts for blends
Separate concepts with :: and optional weights so Midjourney balances elements instead of merging words literally.
Reference images with purpose
Upload references or use style and omni reference parameters when look or likeness matters, and say what should change relative to the source.
Iterate from the best grid cell
Upscale, vary, or open the Editor from the closest result. One change per pass is easier to steer than many simultaneous edits.
Edit with deltas in the Editor
The Midjourney Editor remixes, varies, and edits generated or uploaded images. Describe what changes, not the entire scene again.
Plan for GPU time
Text-to-image uses Fast GPU time included in your plan. Review official plan comparison for monthly allowances, Relax modes, and video limits.
Use personalization when trained
Apply --profile or --p to reuse a custom style you created. Keep scene text simple so the profile can show through.
Discord remains supported
You can also use the /imagine command in Discord. The same prompting and parameter rules apply on web and Discord.
Do not assume a public API
Midjourney does not offer a public API in official docs. Work through midjourney.com or Discord workflows instead.
Review before commercial use
Check Midjourney’s official terms and plan features for commercial rights, privacy settings such as stealth mode, and storage limits on your tier.
- Describe one scene with concrete visual details; avoid long unrelated lists.
- Place parameters at the end with correct spacing; never put prompt text after them.
- Each Imagine run returns four images—pick one and iterate before starting over.
- Use --ar when you know the target shape (feed, slide, poster, portrait).
- Add --no for recurring unwanted elements instead of repeating exclusions in prose.
- Combine text with image references when you need style or subject continuity.
- Check your plan for Fast GPU time, video access, and Relax modes before big batches.
- Use the Editor for localized changes; reserve full re-prompts for large direction shifts.
Try this AI
Product Details
Pricing, features, limits and latest updates
Product Details
Pricing, features, limits and latest updates
Midjourney
AI image and video generation from text prompts and reference images.
Paid · $10 / month
Pricing Plans
Basic
$10/ Monthly
Basic
$96/ Yearly
Standard
$30/ Monthly
Standard
$288/ Yearly
Pro
$60/ Monthly
Pro
$576/ Yearly
Mega
$120/ Monthly
Mega
$1,152/ Yearly
Key Features
Image Generation
Text-to-image on midjourney.com, billed as Fast GPU time included in each plan.
Video Generation
Official plans include video generation; unlimited Relax SD video is on Pro and Mega.
Image Editing
The Midjourney Editor on the website remixes, varies, and edits generated or uploaded images.
Limits
- Basic Fast GPU time: 3.3 hours/month. Included Fast GPU time on the Basic plan; unused time does not roll over (Midjourney plan comparison).
- Standard Fast GPU time: 15 hours/month. Included Fast GPU time on the Standard plan (Midjourney plan comparison).
- Public API: Midjourney documents no public API; automating the bot is prohibited.
Ideas / Prompt experiences
Share a prompt that worked for you. Username and email are shown with your submission. External links are not allowed.
Example prompt
Text-to-image with aspect ratio
Prompt
A quiet street at dusk, wet cobblestones, warm shop lights, shallow depth of field, cinematic photography --ar 16:9
Variation
Same scene, morning light, empty street --ar 16:9 --no rain
Short explanation
Midjourney docs recommend short, clear phrases and parameters such as --ar and --no after the scene text.
Example prompt
Style reference workflow
Prompt
[upload reference image] Product photo of this ceramic mug on a wooden desk, soft window light, minimal background, commercial photography --ar 4:5
Short explanation
Use reference images when you need consistent subject or style; describe what to keep vs change in plain text.
Share your experience
Required fields are marked. Variation, result, and explanation are optional.
You may also like
Runway
ExploreAI image and video generation and editing for motion, storytelling, and creative workflows.
ImageFree / Paid
Ideogram
ExploreImage generation with accurate in-image text, typography, and layout.
ImageFree / Paid
Adobe Firefly
ExploreGenerative AI for images, video, audio, and design built for commercially safe creative work.
ImageFree / Paid