In Midjourney, the prompt weight (or weight parameter) determines how much emphasis is placed on specific words within a text command. I’d love to fully grasp how these weights are calculated in the background and how they influence the output. When assigning different weight values to multiple words, how can we effectively guide the model? How do you explain this concept?
What does the 'prompt weight' concept do in Midjourney and how is it adjusted?
👁️ 156 views💬 2 replies❤️ 0 likes
2 Replies
You actually specify the "prompt weight" using the number of colons after the `:` (e.g., `sunset::2`). MidJourney treats this number as a multiplier that determines how dominant that word or phrase appears in the image. I usually go with values between 1–10; a 1 is subtle, 5 is balanced, and 10 makes that element almost the focal point of the scene.
If you want to assign different weights to multiple words, just add a separate `::weight` to each one. For example: `portrait of a woman ::3, detailed background ::1, neon lights ::5`. Here the woman is mid-priority, the background is subtle, and the neon lights get the most emphasis. Just don’t push the total too high—values above 10–15 can make the model go “over-focused,” so keep the sum in that range. Also, instead of repeating the same word with different weights, just give it the highest value you need; that avoids unwanted clutter.
At the end of the day, it’s all trial and error—run a few tests, tweak the weights, and see what range your model responds to best. And since visual style matters, keeping the global `--stylize` parameter low helps the weighted elements pop more clearly.
Prompt weighting in Midjourney is controlled by assigning values to words using the “::” operator. For example, writing `sunset::2 beach::0.5` makes “sunset” twice as dominant, while “beach” is half as effective. These values range from 0.0 to 2.0, with 1.0 being the default. The weights are automatically normalized and reflected in the model’s attention mechanism, meaning higher-weighted words stand out more in the visual output, while lower-weighted ones stay in the background or as subtle details.
When you want to assign different weights to multiple words, you can use multiple “::” expressions in the same line, but grouping them with parentheses ( ) makes complex combinations easier to manage. For example, `(dragon::1.5 fire::0.8)::1.2` means you first set the ratio between “dragon” and “fire,” then apply an additional weight of 1.2 to the entire group. This way, you can fine-tune both the main theme (dragon) and secondary details (fire) while also controlling the overall emphasis of the group.
What happens when you experiment with negative weights? Or when you combine different variations of the same word in the same prompt (e.g., “old style::1.5” vs. “modern style::0.5”)—how does the model handle the resulting conflicts? Have you tested these scenarios and developed any strategies based on your findings?