Parameters

Stable Diffusion

Weighting Syntax

Stable Diffusion supports both parentheses and numerical weights for emphasis:

Emphasis:
- (keyword) increases emphasis slightly.
- ((keyword)) increases it more significantly.
- (((keyword))) or additional parentheses for even stronger emphasis.
De-emphasis:
- [keyword] reduces the importance of the keyword.
Numerical Weights:
- Use keyword:weight to directly set the weight.
- Example: a cat:1.5, a dog:0.5 gives more prominence to "cat" than "dog."
- Typical weights range from 0.0 (ignore) to 2.0 (high emphasis).

Combining Weighted Terms

Stable Diffusion allows complex prompts with weighted phrases:

Example Prompt:
```
A beautiful ((sunset over the ocean)):1.5, [a city skyline]:0.7, colorful skies:1.2
```
- This prioritizes the sunset while de-emphasizing the city skyline.

Impact of Weighting

Weights directly influence the model’s focus, but the results depend on:

Model Training Data: Some concepts or terms may already be prominent or weak due to the dataset.
Prompt Complexity: Overloading the prompt with high-emphasis terms can dilute results.
Inherent Ambiguity: The model interprets prompts probabilistically, so multiple runs might slightly differ.

Using AND/OR for Advanced Control

You can separate concepts using AND for better multi-object composition:

Example:
```
A futuristic city:1.2 AND a dragon flying in the sky:0.8
```
- This creates a balance between the city and dragon elements

Examples

Detailed Portrait:

A ((realistic portrait)) of a woman with blue eyes:1.3, wearing a red scarf:1.0 --negative (blurry), (cartoonish)

Fantasy Scene:

A magical forest with glowing mushrooms:1.5 AND a fairy with delicate wings:1.2 --negative (dark shadows), (low resolution)

In MidJourney, weighting syntax allows users to emphasize or de-emphasize certain elements in their image prompts. This can guide the AI on which parts of the prompt are more important or should have a greater influence on the final output.

Weighting

Double Colons (::):

Place :: followed by a numerical weight after a specific term or phrase.
Add a space after the numerical weight.
Example: a golden retriever::2 sitting in a field of flowers::0.5
- The golden retriever is emphasized with weight 2 (more influence).
- The field of flowers is de-emphasized with weight 0.5 (less influence).

Examples Prompts

a dragon::1, a castle::1, a knight::1

All elements have equal influence.

a futuristic city::2 a flying car::1 --no skyscrapers

You can use --no with weights to exclude or downplay certain elements. For instance:

This tells MidJourney to focus on the city and car but avoid including skyscrapers.

Combining Image Input with Text Weights

When combining image references and text prompts, the --iw (image weight) parameter adjusts the influence of the uploaded image(s) compared to the text. By default, --iw is 1.

PreviousInstruct Templates NextNegative Prompts

Last updated 6 months ago