Skip to content

Instantly share code, notes, and snippets.

Last active August 8, 2023 16:44
  • Star 14 You must be signed in to star a gist
  • Fork 3 You must be signed in to fork a gist
Star You must be signed in to star a gist
What would you like to do?
midjourney cheat sheet










  • new algo
  • takes significantly longer to run
  • 2 square images or 1 non-square image
  • does not work with --hd, --stylize, and --q 5


  • older algo
  • better for larger images, but with less consistent composition
  • best for abstract and landscape
  • generates images at higher resolutions without the need to upscale

--aspect --ar

  • generates images with the desired aspect ratio (--ar 16:9)

--w --h

  • width and height of the image
  • values used by --h and --w should be between 256 and 2034
  • better as multiples of 64
  • --ar is better supported


  • sets the seed
  • must be a positive integer [0, 4294967295]
  • react with [envelope icon] to a job's message to know what seed value was used


  • the seed affects all images of the resulting grid in the same way


  • negative prompting
  • equivalent to using an advanced text weight of ::-0.5


  • image prompt weight relative to text weight
  • can be negative (e.g. -0.5)
  • default: --iw 0.25

--stylize --s

  • how strong of a 'stylization' your images have
  • the higher you set it, the more opinionated it will be
  • examples:
    • --stylize 625 If you basically want to turn it off and be less artistic.
    • --stylize 1250 Good for when you want it to be 'less strict' but still 'pretty' (this is probably recommended for skilled users).
    • --stylize 2500 The default value, so you don't have to specify this.
    • --stylize 20000 If you want it to 'take over' and start drifting from your text, but not go crazy.
    • --stylize 60000 Hands off the wheels, who knows what will happen. It may look nothing like your prompt.
  • default: --stylize 2500

--quality --q

  • how much rendering quality time you want to spend
  • higher values cost more and lower values cost less
  • possible values (only use one of the values below; any other value will be rounded to a valid value instead):
    • --quality 0.25 Rough results, 4x faster / cheaper.
    • --quality 0.5 Less detailed results but 2x faster / cheaper.
    • --quality 1 The default value, you do not need to specify it.
    • --quality 2 More detailed results, but 2x slower and 2x the price (2 GPU minutes per /imagine).
    • --quality 5 Kind of experimental, might be more creative or detailed... also might be worse! (5 GPU minutes per /imagine).
  • default: 1

--chaos <0-100>

  • how much more varied, random, and different the results will be
  • higher values will favor more interesting and unusual generations in exchange for less reliable compositions


  • faster images, less consistency, less expensive
  • alternative to --q 0.5 and --q 0.25 (similar result)

--stop <10-100>

  • stop the generation at an earlier percentage
  • currently does not work with upscales


  • saves a progress video, which is sent to you in the [envelope icon]-triggered DM
  • you must react with the envelope to get the video link
  • currently does not work with upscales

--v <1,2,3>

  • 1: the oldest model
    • sometimes better for macro or textures
    • formerly the "vibe" option
  • 2: later model
  • 3: current (default)


  • uses the "light" upscaler when selecting the U **** buttons
  • results are closer to the original image, with less detail added during upscale
  • ideal for faces and smooth surfaces

--test --creative --upbeta

  • Artistic enhancement of the image

Text Prompts


Text Weights:

  • hot dog::2 food::-1
    • sends a text prompt of hot dog with the weight 2 and food of weight -1
  • hot dog::0.5 animal::-0.25
    • sends hot dog of weight 0.5 and animal of negative 0.25
  • hot dog:: food::-1 animal::
    • Sends hot dog of weight 1, food of weight -1 and animal of weight 1 (unspecified weights default to 1)
  • hot dog animal::-1
    • sends hot dog animal with weight -1, because of how it parses the text before the colon

Image and Text Weights:

  • [image URL] [text prompt]
    • with no weights specified results in a 20% image(s) / 80% text generation
  • [image URL] [text prompt 1] --iw 1
    • results in a 50% image(s) / 50% text generation
  • [image URL] [text prompt 1]::2 [text prompt 2]::3 --iw 1
    • results in a 1/6=17% image(s), 2/6= 33% text1, 3/6 = 50% text2
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment