The Complete Guide to Prompt Weight Syntax and BREAK

What is prompt weight syntax? The big picture

Weight syntax uses numbers and brackets to specify how strongly a tag should take effect. In Stable Diffusion WebUI (AUTOMATIC1111, reForge, etc.), writing (tag:1.2) multiplies that tag's influence by 1.2. Conversely, (tag:0.8) weakens it, so it's also handy for tuning "tags I want in, but not too pushy."

(tag:1.2) numeric weights: basics and recommended range

The rule of thumb for numeric weights is the 0.8 to 1.4 range. Above 1.5, that one tag often runs wild and the composition or colors start to break down. "It's not working, so let's use 2.0" is risky; usually the real cause is a conflict with another tag. We recommend removing the conflicting tag first, or trying around 1.3 and seeing how it goes.

• 1.1 to 1.2: The safe zone when you want something to stand out a bit
• 1.3 to 1.4: The rough upper limit when you want it to work like a lead element
• 0.8 to 0.9: When you want it in, but subdued
• 0.5 to 0.7: When you want just a hint of it (often doesn't work)
• 1.5 and above: Higher risk of breakdown, so generally don't use it

masterpiece, best quality, 1girl, solo, long hair, (silver hair:1.2), (blue eyes:1.1), school uniform, (cherry blossoms:1.3), outdoors, upper body, smile, looking at viewer, depth of field

Stacked brackets vs. NovelAI {tag}: what's the difference?

Stacking parentheses like ((tag)) gives 1.1 x 1.1 = about 1.21. Triple parentheses give about 1.33. But when you look back later it's easy to lose track of the multiplier, so Rinka sticks to numeric weights like (tag:1.2). Another important point is how NovelAI differs. NovelAI emphasis uses curly braces {tag}, and each pair is designed to be milder than WebUI, about x1.05.

When bringing a prompt from NovelAI to WebUI, reading {{tag}} as roughly equivalent to (tag:1.1) gets the balance close. Pasting curly braces into WebUI as-is won't add any emphasis, so don't forget to rewrite them when you copy a prompt over.

How do you use BREAK to stop concepts from bleeding?

Internally, the prompt is split into 75-token chunks for processing. Writing BREAK (in capitals) forces a chunk boundary at that position. For example, putting BREAK between "character features" and "background settings" reduces bleeding like the hair color seeping into the background. It's a technique worth trying if color bleed is bothering you.

masterpiece, best quality, 1girl, solo, (red hair:1.2), twintails, green eyes, white dress BREAK (blue sky:1.1), sunflower field, summer, wide shot, wind, floating hair

Weights in the negative prompt and fixing breakdowns

Weight syntax works in the negative prompt too. For example, (blurry:1.3) excludes blur more strongly. But overly strong negatives have side effects, such as a stiffer art style or flatter coloring, so keeping them to around 1.2 is safe.

• Keep negative weights within the 0.8 to 1.4 range too
• worst quality, low quality work well enough without any weight
• If three or more emphasized tags are above 1.3, lower some of them
• When the image breaks down, first reset the heaviest tag to 1.0 to isolate the cause

Speed up weight tuning with editor shortcuts

In the AUTOMATIC1111 or reForge prompt box, select the tag you want to adjust and press Ctrl+Up to raise its weight by 0.1, or Ctrl+Down to lower it. It automatically formats it as (tag:1.1), so it's faster than typing and free of typos. ComfyUI also has a setting for similar shortcuts. Rinka's go-to workflow is the loop: generate, select, Ctrl+Up, regenerate. Weight syntax is a foundational skill you'll use forever once you learn it, so start by bringing two things, (tag:1.2) and BREAK, into today's prompts.