Ideal Midjourney Prompt Length for Best Results
Does the length of a prompt matter? Turns out it does. In this post, I will outline how prompt length and image weights factor into Midjourney image generation quality and accuracy.
Understanding Prompt Length in Midjourney
Midjourney operates with a soft maximum of approximately 60 words for prompt length. As the word count increases, different versions of Midjourney demonstrate varying “drop-offs” where words lose influence. For versions --v 1, --v 2, --v 3, --test, and --testp, there’s a linear drop-off as word count approaches 60, with influence gradually decreasing. In version --v 4, the slope of influence is much more forgiving, and words retain their influence longer.
The Permutation Factor
A vital aspect of prompt length is the “combinatorial” or “permutation” factor. Longer prompts increase the number of word permutations, reducing the likelihood that all words will be expressed or expressed in the intended manner. For example:
- With three words, there are eight possible combinations
- With nine words, there are 512 permutations
As the word count increases, the permutations become increasingly vast and unpredictable.
Slope of Influence and Weighted Averages
Midjourney’s slope of influence and weighted averages serve as a substitute for grammar. In version --v 4, natural language processing (NLP) technology helps to enhance the effectiveness of prompt grammar, so the slope isn’t as harsh.
Using Weights and Multiprompts
Using weights or multiprompts adds another layer of complexity. The slope of influence affects both individual word sets (micro perspective) and the entire set of sets (macro perspective). The value of the weight acts as a modifier to the positional influence value, potentially compensating for a word’s late position in the prompt.
Iceberg
iceberg architecture
Mushroom
Mushroom house
Pop Art
Testing
Conclusion
In summary, prompt length and image weights significantly impact the outcome of Midjourney image generations. By understanding the nuances of these factors, users can optimize their prompts to achieve more accurate and aesthetically pleasing image outputs.
Sources:
