ercwang

ercwang

HomeArchives

Deep Dive into Midjourney: Common Issues and Optimization for Prompt (Part 1)

#Midjourney#Prompt优化241
AI Translation
This post is translated from Chinese into English through AI.View Original
AI-generated summary
This document provides tips for improving prompts in Midjourney, a text-to-image generation AI. It suggests using specific adjectives and verbs instead of prepositional phrases, reversing the order of details, and minimizing redundancy. It also recommends iterating prompts multiple times for more accurate results and using higher quality settings. Punctuation is important in prompts, with only certain symbols having programming effects. Other punctuation may add interesting noise but is not necessary.

Translated:

Translated from: this Google document

1. Use the following templates to write better prompts.#

This is not a hard rule, but rather a soft suggestion that many people use to improve prompts.
Break down the prompt into four parts: Subject, Details, Stylizations, and Parameters.

Midjourney --v 5 template

image

Midjourney --v 4 template

image

Midjourney --v 3 template

image

Midjourney --test / --testp template
image

Output of the above example v3

image

Output of the above example test/testp

image

2. Prepositional phrases are unreliable, so try using these alternatives#

✅Use adjective-noun word order instead of prepositional phrases. Replace prepositional phrases with adjectives.

🌐What are prepositional phrases: Refer to this link

For example: (Replace the description on the left with the one on the right)

  • "hair flowing in the wind" ➡️ "flowing hair"
  • "eyes the color of sunset" ➡️ "sunset eyes"
  • "a carrot for a nose" ➡️ "carrot nose"

✅Use more specific verbs instead of prepositional phrases. Replace vague prepositional expressions with more precise actions.
For example: (Replace the description on the left with the one on the right)

  • "a girl with a flashlight" ➡️ "a girl using a flashlight"
  • "a girl with some cake" ➡️ "a girl eating cake"
  • "a girl feeling sad" ➡️ "mourning girl" or "grieving girl" or "depressed girl"
  • "a girl with a big smile on her face" ➡️ "smiling girl" or "happy girl"
  • "a sword on its side" ➡️ "a discarded sword"

✅Use more specific vocabulary instead of prepositional phrases. When you replace vague words with specific ones, you may see better results.
For example: (Replace the description on the left with the one on the right)

  • "a girl in a forest sleeping" ➡️ "a forest shelters a sleeping girl"
  • "a girl walking on a spaceship" ➡️ "a spaceship corridor contains a striding girl"
  • "a girl with a recurve bow" ➡️ "a female ranger equipped with a recurve bow"

✅Try reversing the order of details. This method adjusts the weighting of Midjourney's judgments and may increase coherence.
For example: (Replace the description on the left with the one on the right)

  • "rainbows reflected in the pond" ➡️ "the pond reflects rainbows"
  • "a bee hovers over a flower" ➡️ "a flower visited by a bee"
  • "a storm above a ship" ➡️ "a ship threatened by a storm"
  • "a basket filled with blueberries" ➡️ "blueberries contained by a basket"

✅Minimize redundancy. Check if the prompt already includes the meaning of some adjectives, eliminating unnecessary words can reduce the number of combination results and may increase coherence.

  • "black obsidian" ➡️ "obsidian" (is already black)
  • "botanical plants" ➡️ "botanicals" (are already plants) or "plants" (are already botanicals)

✅Seize every opportunity to increase specificity. Use words with the desired content that include the role/context instead of describing it in a sentence. For example, use "a magician" instead of "a person with a top hat and a magic wand."

3. Iterate the prompt multiple times for more accurate images#

image
Your prompt may need multiple iterations to produce good results. If you have only used /imagine once, your prompt has only generated that one result. "Once" is not effective in a system of combinations and permutations. Choose the closest image from the options and use that image to generate again. After generating multiple times, the results will be more accurate. If you don't like any of the options, scroll through the entire grid again. After 3-5 🔄️ resampling or regenerating, if you still find the results unsatisfactory, it may be necessary to check and restart the prompt text.

4. Quality matters sometimes#

When the --quality command is available, you may consider using a higher value. When you use a higher quality value, you allow Midjourney to "compute" longer in the GPU. It will cost you computation time. The more minutes you spend, the better the output quality. Here, quality refers to technical quality, not aesthetic quality. 😆

5. Punctuation matters, but sometimes it doesn't#

Simply use punctuation in Midjourney according to the usage in natural language. However, if you encounter issues in the prompt, the following information may be helpful for troubleshooting or resolving the problem.

  • Only three types of punctuation have programming effects.

    • Double hyphens separate parameters -- aspect 9:16
    • Double colons separate weighted integers (positive or negative) ::-0.5
    • Spaces are used as separators, which are considered special symbols.

  • Literally, all other punctuation just adds interesting noise.

    • All punctuation adds what we call "interesting noise."
    • Sometimes, comma noise and hyphen noise may be useful.
    • Proper use of commas is recommended to help with grouping, just like in regular writing.
    • Sometimes, hyphens may add relationships between tags when troubleshooting.
Loading...
Ownership of this post data is guaranteed by blockchain and smart contracts to the creator alone.