Regional Prompter Guide: BREAK, ADDCOL, ADDROW + Examples

Regional Prompter splits the image into regions and gives each region its own prompt. It is the usual fix for colors and details bleeding between subjects: red hair ending up on both characters, or the skirt color leaking into the blouse. This page covers the syntax, the settings and ready-to-copy prompts.

Quick start

Install Regional Prompter from the web UI's Extensions → Available tab, restart, and open the Regional Prompter panel under the generation settings.

The README's own first example: three regions stacked top to bottom.

green hair twintail BREAK
red blouse BREAK
blue skirt

Active: On
Use base prompt: Off
Divide mode: Rows
Divide Ratio: 1,1,1
Base Ratio:

Each BREAK starts the prompt for the next region. The first prompt goes to the top third, the second to the middle, the third to the bottom.

Mode names changed. Older tutorials say Horizontal and Vertical. In the current version Horizontal is Columns (regions side by side, left to right) and Vertical is Rows (regions stacked, top to bottom).

Syntax cheat sheet

Divide Ratio accepts plain proportions (1,1,1), fractions (0.1,0.1,0.1 is the same thing) or pixel sizes (300,100,112 for a 512 px side) when you want exact borders.

Copy-paste templates

Two characters side by side

2girls standing in a park, autumn leaves, soft daylight, best quality ADDCOMM
red hair, white sundress, holding a book BREAK
black short hair, denim jacket, holding a coffee cup

Divide mode: Columns
Divide Ratio: 1,1

ADDCOMM sends the first line to both regions, so each half knows the scene. With common on you need one more prompt than regions: here 3 prompts for 2 regions.

Outfit from top to bottom

1girl, full body, standing, plain studio background ADDCOMM
silver hair, blue eyes BREAK
black leather jacket, white t-shirt BREAK
red plaid skirt, black boots

Divide mode: Rows
Divide Ratio: 1,1,1

Sky, mountains, lake

landscape photo, golden hour, high detail ADDCOMM
dramatic orange clouds, sunset sky BREAK
snowy mountain range, pine forest BREAK
calm lake, reflections

Divide mode: Rows
Divide Ratio: 2,2,1

Base prompt instead of common

a cozy living room, warm evening light ADDBASE
blue velvet sofa BREAK
green plants by a large window

Divide mode: Columns
Divide Ratio: 1,1
Base Ratio: 0.2

Each region is generated as 20% "cozy living room" + 80% its own prompt. With base on you also need one extra prompt: 3 prompts for 2 regions.

Per-region negative prompts

Negative prompts can be split with BREAK too. If the negative prompt has no BREAK, the same negative is used for every region.

blurry, lowres BREAK
blurry, lowres, extra arms

For ready-made negative lists, see the negative prompts guide.

Common vs base

Both share text between regions, but in different ways:

You can use both. The order is: common, base, then the region prompts. Three regions with both on need five prompts.

2D regions with ADDCOL and ADDROW

For grids, use ADDCOL and ADDROW instead of BREAK. Prompts are read top to bottom, left to right. The README's example:

(blue sky:1.2) ADDCOL
green hair twintail ADDCOL
(aquarium:1.3) ADDROW
(messy desk:1.2) ADDCOL
orange dress and sofa

Divide mode: Columns
Divide Ratio: 1,2,1,1;2,4,6

How to read the ratio: groups are separated by ;, one group per row. The first number in a group is the row's height, the rest are the column widths in that row. So 1,2,1,1 is a row of height 1 with three columns (2:1:1), and 2,4,6 is a row of height 2 with two columns (4:6). That is 3 + 2 = 5 regions, matching 5 prompts.

Another grid from the README, with a common prompt:

fantasy ADDCOMM
sky ADDROW
castle ADDROW
street stalls ADDCOL
2girls eating and walking on street ADDCOL
street stalls

The extension can draw the regions for a ratio and generate a matching prompt template for you: enter the ratio and press the visualize button. Use it whenever you are unsure how a ratio splits.

If the ratio does not match the number of separators, all regions are made equal. The flip option swaps , and ;, which rotates the layout by 90 degrees.

Attention vs Latent mode

• Attention (default): use it normally. Fast, but at 512×512 regions are reduced to about 8×8 pixels inside the U-Net, so small regions mix together.
• Latent: slower (generation time × number of regions) but keeps regions more exact, and lets you apply different LoRAs to different regions to some extent.

You can write BREAK or AND in either mode; the extension converts between them.

Support by front end, from the README:

The 2026-09-04 update added Forge Neo models (Z-Image, Krea2, Anima): Latent mode works on all three, Attention works on Krea2 and Anima (on Z-Image it can merge two central subjects), and Region LoRA does not work on them.

Mask and Prompt modes

• Mask: draw the regions by hand (or upload a mask image) instead of using ratios. Divide Ratio is ignored, Base Ratio still applies. Handy for odd shapes; trace over an OpenPose or depth image if you want.
• Prompt: the region is found from a word in the prompt. Example from the README, putting the apple print only on the shirt:

lady smiling and sitting, twintails green hair, white skirt, shirt BREAK
(apple printed:1.4),shirt

The threshold setting controls how big the detected area is; use one value per target, separated by commas (for example 0.4,0.6). Strong weights such as 1.4 to 1.6 on the target effect are normal here.

Troubleshooting

Regional prompting in SwarmUI and ComfyUI

SwarmUI has region syntax built into the prompt:

a photo of a cat and a dog on a sofa <region:0,0,0.5,1> a ginger cat <region:0.5,0,0.5,1> a black labrador

The format is <region:x,y,width,height,strength>, all values as fractions of the image (0.5 = half). Strength is optional. <region:background> targets everything not covered by another region, and <object:...> adds an automatic inpaint pass over the region. SwarmUI's docs note that SDXL follows regions strongly, while SD3 and Flux tend to apply them only in early steps, and very small regions are likely to be ignored.

ComfyUI does it with nodes: the official Area Composition examples use the core ConditioningSetArea node to give each area of the image its own prompt. They include a 16:9 image with a square area for the subject, which keeps wide images more coherent.

Related guides

To add random variation inside regions, combine Regional Prompter with wildcards. For other extensions that work alongside it, see the Stable Diffusion extensions guide. For weight syntax such as (word:1.3), see the prompt weighting guide.

Build your prompt

The prompt generator builds the description for each region from subject, style and lighting choices. Paste the parts between BREAKs. It runs in your browser and does not generate images.

Primary sources

• Regional Prompter README (hako-mikan/sd-webui-regional-prompter)
• AUTOMATIC1111 wiki: Extensions
• SwarmUI docs: Prompt Syntax (Regional Prompting)
• ComfyUI examples: Area Composition