If you want more consistent outputs, the key is to write Midjourney prompts “clearly and reusably.” This tutorial goes from zero to practical use, explaining the basic structure of Midjourney prompts, how to add reference images, how to combine common parameters, and how to iterate quickly without making things messier with every change.
1. First, write Midjourney prompts as “readable sentences”
The most convenient Midjourney prompt structure is: subject + key features + scene/lighting + style + image-quality description. Starting the subject with a noun is more reliable, such as “a ceramic teapot / 一把陶瓷茶壶,” then add clear features like material, color, a sense of age, and so on.
Be as specific as possible. Use fewer subjective words like “good-looking, exquisite, stunning,” and replace them with visualizable terms such as “soft rim light, shallow depth of field, matte texture.” When writing Midjourney prompts, first make sure the information doesn’t conflict, then consider adding style terms.
2. How to use reference images: use links + weights to control direction
Paste reference image links at the very beginning of the Midjourney prompt (the images must be publicly accessible) to “anchor” the composition, color palette, or the subject’s form. If you find the result is being pulled off track by the reference image, you can lower the image weight (for example with --iw) so the text description takes the lead.
A practical habit is: use one reference image to lock composition, then use text to clearly specify materials and details; don’t throw in too many reference images at once, or the styles will clash with each other. If there are elements in the Midjourney prompt that “must be kept,” repeat them and place them closer to the front.


