Prompting: Weights

When generating AI images, weights play a major role in shaping the final outcome. Weights are present in checkpoints, LoRAs, and, of course, your prompt.

This guide focuses on how to control weights effectively to ensure your desired elements appear in the generated image.

Have you ever wondered why some parts of your prompt won’t show up? A weight shift may be the reason.

1. Applying Weight

As mentioned in an earlier guide, you can increase the weight of a specific part of your prompt by adding more details.

Example:

📌 Base Prompt:
"Realistic photo of a beautiful brunette woman wearing a hat, sitting pose, train interior background."

If the hat doesn’t appear, you can reinforce it by adding more detail:

📌 Revised Prompt:
"Realistic photo of a beautiful brunette woman wearing a sophisticated red hat, deep-seated above the eyeline, sitting pose, train interior background."

This approach makes the hat more likely to appear but reduces flexibility, as the AI now generates a specific kind of hat.

Alternative Methods to Control Weight:

a) Mathematical Approach (Round Brackets) 🔢

Increase an element's importance by adding a numerical weight (from 0 to 1.5):
📌 "Realistic photo of a beautiful brunette woman wearing a (hat:1.5), sitting pose, train interior background."

b) Weight Reduction in SD 1.5 (Square Brackets) ⬜

Use square brackets to reduce the importance of certain elements:
📌 "[Realistic photo of a beautiful brunette woman] wearing a hat, [sitting pose, train interior background]."

c) Repetition 📣

Repeating an element reinforces its significance:
📌 "Realistic hat magazine photo of a beautiful brunette woman wearing a hat, sitting pose, train interior background."

d) Emphasizing with CAPITALIZATION 🔠

📌 "Realistic photo of a beautiful brunette woman wearing a HAT, sitting pose, train interior background."

e) Emphasizing with Exclamation Marks ❗❗❗

📌 "Realistic photo of a beautiful brunette woman wearing a !!!hat!!!, sitting pose, train interior background."

2. Using Weights to Mix Elements (Great for SD 1.5)

In Stable Diffusion 1.5, you can mix different concepts using vertical bars ( | ) in brackets:

📌 Simple 50/50 Mix:
📌 "[Thing1|Thing2]"

📌 Fine-Tuning the Mix:
📌 "[Thing1:0.2|Thing2:0.8]" → Adjust the weight ratio between the two elements.

📌 More Complex Mixes:
📌 "[Thing1|Thing2|Thing3]"

Summary

Different approaches allow you to fine-tune weights and increase the chances of generating the exact image you want. Experimenting with these methods will help you master AI prompting and achieve better control over your outputs. 🚀