# Prompt weighting Sometimes the model ignores a crucial part of your prompt or focuses too much on a minor detail. **Prompt Weighting** allows you to manually adjust the importance of each token. ### Syntax (Automatic1111 / ComfyUI) Most interfaces use parenthesis for emphasis: * `(word)` increases attention by a factor (usually 1.1x). * `[word]` decreases attention by a factor (usually 0.9x). * `(word:1.5)` or `(word:0.8)` explicitly sets the weight to 1.5x or 0.8x. > [!example] > - `a photo of a cat` > -> The model balances "photo" and "cat" **equally**. > - `a photo of a (cat:1.3)` > -> The model prioritizes "cat" features **more strongly**. > - `a photo of a [cat]` > -> The cat becomes **less prominent**, perhaps blending into the background. ### Best Practices * **Keep it Subtle**: Weights between `0.8` and `1.2` usually suffice. Going above `1.5` often breaks the image, causing artifacts or "deep-frying" colors. * **Fix Bleeding**: If `red shoes` is making your whole image red, try `(red shoes:0.8)` or move it later in the prompt. * **Emphasize Style**: Weighting stylistic keywords (e.g., `(oil painting:1.2)`) is a great way to force a specific look without changing the subject. ### Use Cases * **Prioritize the Subject**: When competing concepts fight for attention (e.g., "red dress" and "winter scene"), weight the more important one: `(red dress:1.3) winter scene`. * **Tone Down Unintended Effects**: If a style modifier is too strong, reduce it: `(cyberpunk:0.8)` instead of fully removing it. * **Layer Rare Concepts**: Uncommon phrases ("art nouveau," "chiaroscuro") need higher weights to break through and affect the image: `(chiaroscuro:1.5)`. * **Fix Lighting Issues**: If your prompt lacks lighting control, weight lighting terms: `(volumetric lighting:1.2) (golden hour:1.1)`.