Skip to content

IA.md

Latest commit

 

History

History
225 lines (183 loc) · 10.7 KB

IA.md

File metadata and controls

225 lines (183 loc) · 10.7 KB
brand Midjourney
tagline An independent research lab exploring new mediums of thought.
category AI & ML
website https://midjourney.com

Midjourney — Information Architecture

Overview

Midjourney is an AI image generation platform known for its distinctive aesthetic and artistic quality. Originally operating exclusively through Discord, the product has expanded to a dedicated web application at midjourney.com. The IA reflects this unique evolution: a gallery-centric creation workspace where the primary interaction loop is prompt → generation → variation → upscale → organize. The web app centers on a visual grid of generated images, with a persistent prompt bar at the bottom. Key features include the prompt editor, image editor (inpainting/outpainting), style references, character references, and organizational features like folders and image ratings. The Discord-native /imagine command remains a parallel interface.

Site Map

midjourney.com
├── / (Home / image feed)
├── /explore (Community showcase)
├── /create (Generation workspace)
├── /imagine (Prompt editor with advanced params)
├── /archive
│   ├── / (All generated images)
│   ├── /likes (Liked images)
│   ├── /folders (Organized folders)
│   └── /folders/{folder_id}
├── /image/{image_id} (Individual image detail)
├── /editor/{task_id} (Image editor — inpaint/outpaint)
├── /profile/{username} (Public profile gallery)
├── /account
│   ├── /billing
│   └── /settings
├── /docs (Documentation)
└── /auth

Navigation Model

  • Primary navigation: Top bar with tabs — Home, Explore, Create, Archive
  • Persistent element: Bottom prompt bar (always visible in Create mode)
  • Secondary navigation: Archive sub-tabs — All, Likes, Folders, Smart Folders
  • Image navigation: Click any image → full-screen lightbox with action buttons (Vary, Upscale, Remix, Edit)
  • Utility navigation: Top-right profile avatar → Account, Billing, Settings, Log out
  • Filter navigation: Explore page has style/tag filters; Archive has date/model/aspect ratio filters
  • Mobile: Bottom tab bar replaces top nav; prompt bar adapts to keyboard

Content Model

Entity Key Attributes Relationships
Image Generation (Job) 4-image grid from a single prompt, with model version, parameters, seed User-owned
Individual Image High-res image with metadata: prompt, params (--ar, --v, --s, --c), seed, model Part of job
Variation New generation derived from a parent image with modified prompt/params User-owned, linked to parent
Upscale Enhanced resolution version of a selected image User-owned
Folder Named container for organizing images User-owned
Style Reference Image used as visual style input for new generations Linked to job
Prompt Text description + parameters (aspect ratio, stylize, chaos, weird, model version) Part of job

User Flows

Basic Image Generation

Navigates to Create → focus on prompt bar at bottom → Types descriptive prompt (e.g., "a cyberpunk cat in neon rain, cinematic lighting") → Adjusts parameters: aspect ratio, model version (v6, niji), stylize, chaos → Submits → 4-image grid generates (30-60 seconds) → Selects favorite → Upscale (U1-U4) or Vary (V1-V4, Subtle/Strong) → Final image saved to Archive

Image Editing (Inpaint/Outpaint)

Selects an image → clicks Edit → Editor opens with brush tools for masking regions → Masks area to change → types replacement prompt → Midjourney regenerates masked area while preserving the rest → Can also extend canvas (outpaint) in any direction

Style/Character References

Attaches an image URL as `--sref` (style reference) or `--cref` (character reference) → Midjourney extracts visual style or character features → New generation applies the referenced style/character to the new prompt → Enables consistent character design across multiple images

URL / Route Structure

Pattern Description
/ Home feed
/explore Community gallery
/create Generation workspace
/archive User's image archive
/archive/folders/{uuid} Specific folder
/image/{uuid} Image detail page
/editor/{uuid} Image editor
/profile/{username} Public user profile
/account/* Account management
/docs/* Documentation

Image IDs are UUIDs. Public profiles use usernames. No SEO slugs on images (visual content, not text).

Search & Filter

  • Prompt search: Search own Archive by prompt text
  • Explore search: Search community images by keyword/prompt
  • Filters: Model version (v5, v6, niji), aspect ratio, date range, favorites
  • Smart Folders: Auto-organized by AI-detected content categories
  • Rating system: Users can rate own images (1-5 stars) to train personal style preferences
  • No tag system: Organization relies on prompt text matching and folders

Responsive Behavior

Breakpoint Behavior
Desktop (>1280px) Full grid view (4-6 columns), persistent prompt bar, side panels for image detail
Desktop (1024-1280px) 3-4 column grid, prompt bar at bottom
Tablet (768-1024px) 2-3 column grid, prompt bar at bottom, image detail as overlay
Mobile (<768px) 1-2 column grid, prompt bar fixed at bottom above keyboard, image detail full-screen
  • Image grids use masonry layout adapting to viewport width
  • Prompt bar is always fixed to bottom, expands on focus to show parameter controls
  • Lightbox view is full-screen on mobile with swipe-to-navigate
  • Touch: pinch-to-zoom on images, swipe between images in lightbox

Access Control

Role Permissions
Anonymous Browse Explore gallery only; no generation
Free Trial ~25 free generations to try the service
Basic ($10/mo) ~200 generations/mo, standard queue, community gallery
Standard ($30/mo) ~900 generations/mo, fast hours, unlimited relaxed, stealth mode option
Pro ($60/mo) ~1800 generations/mo, more fast hours, stealth mode, concurrent jobs
Mega ($120/mo) ~3600 generations/mo, maximum fast hours, 12 concurrent jobs
  • Authentication: Discord OAuth (primary), email/password (web app)
  • Image visibility: Public by default; Stealth mode (Pro+) makes images private
  • Commercial usage: All paid plans include commercial license
  • Content moderation: NSFW content blocked; prompts filtered by content policy
  • Discord integration: All web generations also appear in Discord DMs (if linked)

Prompt Parameters

Parameter Syntax Description
Aspect Ratio --ar 16:9 Output dimensions (1:1, 4:3, 16:9, 9:16, etc.)
Model Version --v 6 Generation model (v5, v5.2, v6, niji 6)
Stylize --s 250 How strongly Midjourney applies its aesthetic (0-1000)
Chaos --c 50 Variation in initial grid results (0-100)
Weird --w 500 Experimental/unconventional aesthetics (0-3000)
Quality --q 1 Rendering time vs quality tradeoff (0.25, 0.5, 1, 2)
Tile --tile Generate seamless repeating patterns
No --no trees Negative prompt — exclude elements
Style Reference --sref URL Apply visual style from reference image
Character Reference --cref URL Maintain character consistency across images
Seed --seed 12345 Reproduce similar results with same seed
Stop --stop 80 Stop generation partway for abstract/incomplete look

Generation Pipeline

Prompt Input → Text Encoding → Diffusion Model Processing → 4-Image Grid Output → Select Image → Upscale (2x/4x) or Vary (Subtle/Strong) → Final High-Res Image → Save to Archive

Image Actions

  • Upscale: Increase resolution (Subtle: preserve detail, Creative: add detail)
  • Vary (Subtle): Minor variations keeping core composition
  • Vary (Strong): Significant reinterpretation of the image
  • Vary (Region): Inpaint — regenerate a selected area with new prompt
  • Zoom Out: Extend canvas outward with AI-generated surroundings
  • Pan: Extend canvas in a specific direction (left/right/up/down)
  • Remix: Re-run with a modified prompt while keeping composition

Model Comparison

Model Strengths Release
v5 Photorealism, hands, text accuracy Mar 2023
v5.2 Better aesthetics, higher variation Jun 2023
v6 Improved prompt understanding, text in images, coherent scenes Dec 2023
Niji 6 Anime/manga style, character consistency Dec 2023

Discord Integration

  • /imagine — Generate from prompt in any channel or DM
  • /describe — Upload image, get prompt suggestions
  • /blend — Combine 2-5 images into one
  • /shorten — Analyze prompt and suggest shorter version
  • /settings — Configure default model, quality, style
  • Reactions: Click U1-U4 to upscale, V1-V4 to vary (on grid images)
  • Direct Messages: Private generation (all plans)
  • Community channels: See others' generations in real-time for inspiration

Community & Discovery

  • Explore feed: Curated community gallery sorted by hot/top/new
  • Profile pages: Public gallery of user's generations (unless Stealth mode)
  • Trending styles: Discover popular aesthetic movements and techniques
  • Image details: Click any public image to see its prompt and parameters
  • Reuse prompt: Copy another user's prompt as starting point
  • Rate images: Star system for personal organization and style training
  • Daily theme: Community prompt challenges

Content Policy

  • Banned content: Gore, NSFW, hateful imagery, real person deepfakes
  • Prompt filtering: Automated word/phrase blocking with manual review
  • Content moderation: Community reporting + automated detection
  • Terms of Service: All paid users get commercial license for generated images
  • Stealth mode: Pro+ users can make generations private (not on Explore feed)

Pricing Deep Dive

Feature Basic $10 Standard $30 Pro $60 Mega $120
Fast GPU min/mo 3.3 hrs 15 hrs 30 hrs 60 hrs
Relaxed mode — Unlimited Unlimited Unlimited
Concurrent jobs 3 3 12 12
Stealth mode — — ✅ ✅
Max upscale 2x 2x 4x 4x

Web vs Discord

Feature Web App Discord
Image generation ✅ ✅
Image editing ✅ —
Archive management ✅ —
Folders ✅ —
Style reference ✅ ✅
Describe — ✅
Blend — ✅
Community gallery ✅ —