Prompt Prioritization & Weighting

Fine-tuning prompts for more precise control

Word Position → Higher Weight

The order of your prompt terms affects the generated image. Terms at the beginning of the prompt generally receive more attention from the AI model.

Beginning of the Prompt

A portrait of a woman

with a red dress and a necklace, oil painting, detailed, ornate

→ Focus is on the person (woman), dress and necklace are secondary

End of the Prompt

Oil painting, detailed, ornate, with a red dress and a necklace,

a portrait of a woman

→ Style elements might be more dominant than the main subject

Basic Rule:

Place the main subject at the beginning of the prompt and stylistic details towards the end.

Recommended Prioritization Order:

  1. Main Subject — What should be the focus?
  2. Composition/Environment — How is the scene structured?
  3. Important Attributes — Specific properties of the subject
  4. Artistic Style — What aesthetic should the image have?
  5. Quality Markers — Sharpness, detail level, resolution, etc.

Weighting Syntax: ( ), [ ] and :n

Using special syntax, you can give specific terms or phrases more or less weight, regardless of their position.

Syntax Weighting Example Effect
(Term) 1.1× (portrait of a woman) Slight emphasis
((Term)) 1.21× (1.1²) ((portrait of a woman)) Medium emphasis
(((Term))) 1.33× (1.1³) (((portrait of a woman))) Strong emphasis
[Term] 0.9× [portrait of a woman] Slight de-emphasis
[[Term]] 0.81× (0.9²) [[portrait of a woman]] Medium de-emphasis
Term:1.5 1.5× portrait of a woman:1.5 Exact weighting

Visual Impact of Weighting

Portrait with normal prompt weighting

Normal Prompt

Portrait with double emphasis weighting

Double Emphasis

Portrait with de-emphasized weighting

De-emphasized

Practical Tips:

  • Don't overdo it: Too many weights make the prompt unreadable
  • Experiment: Test different weights with the same seed
  • Be selective: Only weight the truly important elements
  • Document: Note successful weightings for later use

Balancing Topics: Blend / AND / OR

For complex prompts with multiple subjects or styles, you can use special techniques to combine or separate them.

Blend Technique

a house AND a castle

Creates a mix of both elements

OR Technique

a house | a castle

Generates either one or the other

AND with Weighting

a house:1.2 AND a castle:0.8

Controlled mix with dominance

Example Applications

Architectural Style Mix

modern architecture:1.2 AND gothic cathedral:0.8, detailed, atmospheric, sunlight

Architectural mix of modern and gothic styles

Alternative Generation

a portrait in the style of picasso | a portrait in the style of van gogh, detailed, artwork

Portrait in Picasso style

Variant 1

Portrait in Van Gogh style

Variant 2

Compatibility Note:

The AND/OR syntax does not work the same way in all UI implementations. In A1111 WebUI, "AND" (uppercase) is used, while ComfyUI might use different notation.

Practical Exercise

Exercise: Weighting and Balance

Try the following prompt variations with the same seed and compare the results:

Base Prompt

portrait of a woman with red jacket, city in background, sunlight, detailed

With Weighting

((portrait of a woman)) with red jacket, [city in background], sunlight, detailed

With AND Technique

portrait of a woman with red jacket AND city in background, sunlight, detailed

Learning objective achieved:

You now understand how word position and weighting syntax direct the model's attention and how you can balance multiple topics in your prompts.