Install ComfyUI.
Install ComfyUI-Manager
Look up this extension in ComfyUI-Manager. If you are installing manually, clone this repository under ComfyUI/custom_nodes.
Restart ComfyUI.

Nodes

PromptPresetSelector

PromptPresetSelector v2.1

Inputs

syntax (str), default to {% | %} — defines the opening tag, separator, and closing tag as a single field, space-separated
preset_index (int), 0-based, default to 0 — loops both ways within range
preset_name (combo) — dynamic dropdown, active when preset_names is filled
preset_names (str), optional — comma-separated list of preset names. Ex: neutral, happy, sad. If fewer names than presets, remaining are auto-named (preset_2, preset_3...). If more, extra names are ignored.
text (str), multiline textarea supporting preset blocks
cleanup (bool), default to True

Outputs

text (str) — the formatted text, with preset blocks resolved
preset_index (int) — the selected index
preset_count (int) — the number of presets found across blocks

Usage

Include preset blocks in your text using the syntax defined in the syntax field. Each block holds options separated by the separator. Depending on preset_index, the matching option is injected and the block markers are removed.

This is {% an example | a text | another text %} for {% the PromptPresetSelector | the README | %}

preset_index = 0 → This is an example for the PromptPresetSelector
preset_index = 1 → This is a text for the README
preset_index = 2 → This is another text for

Empty presets are allowed — the block is simply replaced by an empty string.

Custom syntax

The syntax field allows changing the opening tag, separator and closing tag. Format is always open_tag separator close_tag, space-separated.

<< / >>

Would match blocks like << option_a / option_b >>.

Preset names

When preset_names is filled, preset_index is replaced by a dynamic dropdown (preset_name) built from the comma-separated names. The dropdown resets to the first preset when activated.

If fewer names than presets are defined, remaining entries are auto-named (preset_2, preset_3...). If more names than presets, extra names are ignored.

Preset references

Inside a preset, you can reference a previously declared preset using $N (0-based index).

{% base style | $0, hat | $1, jewelry %}

preset_index = 0 → base style
preset_index = 1 → base style, hat
preset_index = 2 → base style, hat, jewelry

References are resolved recursively — $1 in preset 2 resolves preset 1, which may itself reference preset 0.

Rules:

$N can only reference a preset with a lower index (forward references are not allowed)
$N must be within range
Circular references raise an error

Cleanup

When cleanup is enabled, the following are removed from the output:

Spaces before commas
Multiple consecutive spaces
Empty or comma-only parentheses and brackets
Residual commas after opening or before closing parentheses
Double or empty commas
Leading commas
Empty lines

Validation

The node runs the following checks and stops the workflow on error:

syntax must contain exactly 3 space-separated parts
open_tag and close_tag must be different
All preset blocks in text must have the same number of options
preset_index must be within range

PromptOptionPicker

PromptOptionPicker_PromptLayoutFiller_v1

Inputs

options (str) — one option per line. Empty lines are allowed and displayed as -- in the dropdown, with an empty string as value.
selected (combo) — dynamic dropdown built from the options field.

Outputs

value (str) — the selected option, or "" if -- is selected.

Usage

Fill the options field with one option per line. The dropdown updates automatically as you type.

sitting down
standing up
lying down

kneeling

The empty line above produces a -- entry in the dropdown, which outputs an empty string. If the field is empty, the dropdown resets to --.

Labels

Options can be labeled using the label $: value syntax. The label is shown in the dropdown, but the real value is transmitted on output:

neutral $: character, simple background
happy $: character, smile, laughing
sad $: character, crying

For unlabeled options, use $: value — the option is auto-named option_N:

$: character, simple background
$: character, smile

Cartesian product blocks

Use @combine / --- / @end blocks to generate all combinations of multiple lists:

@combine
from front
straight-on
from side
---
from above
from below
---
close-up
full body
@end

Each --- separates a new list. The block expands into all combinations joined by , :

from front, from above, close-up
from front, from above, full body
from front, from below, close-up
...

Multiple @combine blocks are allowed and processed independently. A --- outside a block is treated as a literal option.

Validation

The node stops the workflow on error if:

A @combine block is opened inside another
A @end is found without a matching @combine
A @combine block is never closed

PromptLayoutFiller

PromptOptionPicker_PromptLayoutFiller_v1

Inputs

template (str) — multiline text with placeholders. Supports {N} or {N:label} syntax where N is the slot index (0-based) and label is a purely visual annotation.
cleanup (bool), default True
slot_0 to slot_9 (str, connectable) — values injected into the corresponding placeholders. Slots are revealed one by one as the previous one is connected. Up to 10 slots supported.

Outputs

text (str) — the template with all placeholders resolved.

Usage

Write a template in the text field using placeholders:

character, {0:position}, {1:background}, {2:lighting}

Connect a node (e.g. a PromptOptionPicker) to each slot. The placeholder is replaced by the connected value at execution.

Presets

The node supports saving and recalling slot configurations as named presets, stored directly in the workflow.

[preset dropdown] [➕] [💾] [🗑️]
[📋 Export]  [📥 Import]

| Button | Action | | ---------- | ------------------------------------------------------------------ | | ➕ | Creates a new preset from current slot values (prompts for a name) | | 💾 | Overwrites the selected preset with current slot values | | 🗑️ | Deletes the selected preset and recalls the next one if available | | 📋 Export | Copies all presets as JSON to clipboard | | 📥 Import | Pastes and loads presets from JSON |

Presets are sorted alphabetically. Recalling a preset pushes the saved values into connected source nodes. If a saved value is no longer available in a source node, or the source node is disconnected, an error is raised.

Cleanup

Same rules as PromptPresetSelector. Useful when a connected slot outputs an empty string, leaving residual commas or parentheses.

Validation

The node stops the workflow on error if:

The template is empty
A placeholder index exceeds the maximum slot index (9)
A placeholder is used but the corresponding slot is not connected

Typical workflow

[PromptOptionPicker] → slot_0 ┐
[PromptOptionPicker] → slot_1 ├→ [PromptLayoutFiller] → text
[PromptOptionPicker] → slot_2 ┘

Each PromptOptionPicker manages its own list of options independently. Slot lengths do not need to match.

PromptSwitch

PromptSwitch_v1

Inputs

on_true (str) — value returned when condition is True. Accepts direct input or connected node.
on_false (str) — value returned when condition is False. Accepts direct input or connected node.
condition (bool), default True — controls which input is routed to the output.

Outputs

text (str) — the selected value based on condition.

Usage

Connect or type a value in on_true and on_false. Toggle condition to switch between them.

If the node is bypassed, an empty string is returned.

Editor

Input ports and widget borders are colored based on the active condition:

Orange: active input when condition is True
Blue: active input when condition is False
Gray: inactive input Colors are only shown when an input is connected or has a value.

PromptRouter

PromptRouter_v1

Inputs

selected (str, internal) — stores the active selection for workflow persistence.
_sub_selected (str, internal) — stores the active sub-router selection.
input_0 to input_9 (str, connectable) — values to route. Slots are revealed one by one as the previous one is connected. Up to 10 slots supported.

Outputs

text (str) — value of the selected input.
index (int) — index of the selected input.

Usage

Connect nodes to the input slots. The dropdown automatically populates with the title of each connected source node, prefixed by its index:

0: position
1: background
2: lighting

Select the active input from the dropdown. The node outputs the value of the selected input.

If a connected node is bypassed, the output is an empty string.

Sub-router

If a connected source node is itself a PromptRouter, it is marked with a ▶ indicator in the dropdown:

0: position
1: Router B ▶
2: background

Selecting it reveals a second dropdown populated with the options of the child router. The parent outputs the value selected in the child dropdown.

Changing the selection in either the parent sub-dropdown or the child router updates both in sync. Multiple parents connected to the same child router all stay synchronized.

Sub-routing is limited to one level deep.

Special node integration

When a connected source node is a PromptOptionPicker, PromptLayoutFiller, or PromptPresetSelector, it is marked with a ◆ indicator in the dropdown:

0: position ◆
1: Router B ▶
2: background

Selecting it reveals a third dropdown populated with the options or presets of that node. The selection is synchronized in both directions: changing the value in the Router updates the source node, and changing it in the source node updates the Router.

This also works one level deep through a sub-router:

[Router] → [sub-router ▶] → [OptionPicker ◆] → [value dropdown]

Editor

The dropdown updates automatically when source nodes are connected, disconnected, or renamed.
The active selection is preserved across workflow saves, reloads, and undo/redo.

Known limitations

If a child router's connections change while it is selected in the parent, the parent dropdown may not update immediately. Reconnecting or modifying the parent will trigger a refresh.
If a getter node is connected to the router, the getter's name will be in the router's dropdown but, the getter being "invisible" (directly getting the setter's value), the value would be provided by the node before the setter. Thus, the router would receive an empty string, because of not getting the right node. Linking the getter with a string node before connecting to the router solves the issue.

QuickCombo

Inputs

options (str) — comma-separated list of options typed inline. Ex: neutral, happy, sad
selected (combo) — dynamic dropdown built from the options field
one_based (bool), default False — if True, the index output starts at 1 instead of 0

Outputs

value (str) — selected option
index (int) — index of the selected option (0-based by default)

Usage

Type a comma-separated list in the options field. The dropdown updates automatically as you type.

neutral, happy, sad, angry

Toggle one_based to match the indexing convention of other nodes in your workflow.

Differences from PromptOptionPicker

| | QuickCombo | PromptOptionPicker | | ----------------- | ---------------------- | ----------------------- | | Input format | Inline comma-separated | Multiline, one per line | | @combine blocks | No | Yes | | Labels ($:) | No | Yes | | Empty options | No | Yes (--) | | Use case | Quick setup | Advanced lists |

StyleLoader

StyleLoader_v1

Inputs

style_file (combo) — dropdown listing all .json files found recursively under the styles/ folder. Supports subfolders (e.g. anime/illustrious.json).
checkpoint (combo) — model checkpoint to load.
vae (combo) — VAE override. Select none to use the checkpoint's built-in VAE.
clip_skip (int), default -2 — number of CLIP layers to skip. Set to -1 to disable.
quality_tags (str) — positive quality prompt, passed through as a string output.
extra_quality_tags (str) — extra positive quality prompt, passed through as a string output.
negative_tags (str) — negative quality prompt, passed through as a string output.
steps (int), default 15 — number of sampling steps.
refiner_step (int), default 24 — step at which a refiner pass begins, if used.
cfg (float), default 4.0 — classifier-free guidance scale.
sampler (combo) — sampler name, sourced from ComfyUI's own sampler list.
scheduler (combo) — scheduler name, sourced from ComfyUI's own scheduler list.
loras_data (str, internal) — serialized LoRA list for workflow persistence.

Outputs

model — loaded model with LoRAs applied.
clip — CLIP encoder with clip skip and LoRAs applied.
vae — VAE (overridden or from checkpoint).
quality_tags (str) — positive prompt passthrough.
extra_quality_tags (str) — extra positive prompt passthrough.
negative_tags (str) — negative prompt passthrough.
steps (int)
refiner_step (int)
cfg (float)
sampler (str)
scheduler (str)

Usage

Style files are .json files stored under the src/data/styles/ folder at the root of the extension. Subfolders are supported and reflected in the dropdown after a page refresh.

A style file contains all parameters for a generation setup:

{
  "checkpoint": "model.safetensors",
  "vae": "none",
  "clip_skip": -2,
  "loras": [
    { "name": "lora.safetensors", "weight": 0.8 }
  ],
  "quality_tags": "masterpiece, best quality",
  "extra_quality_tags": "perfect face",
  "negative_tags": "worst quality, low quality",
  "steps": 15,
  "refiner_step": 24,
  "cfg": 4.0,
  "sampler": "euler_ancestral",
  "scheduler": "karras"
}

LoRAs support a single weight (applied to both model and clip), or separate model_weight and clip_weight.

Preview

You can embed a preview image for each of your styles by putting an image with the same name as the style in its directory. The node looks for png, jpg, webp or jpeg files, and display the image if found on top of the style_file input, next to the outputs.

Editor

The node includes a built-in styled editor:

All fields are native ComfyUI widgets — checkpoints, VAE and samplers have built-in search and autocomplete.
LoRA rows are displayed as canvas widgets, one per line, with a name picker, weight control, and remove button.
The weight can be adjusted by dragging horizontally, clicking the ◀ / ▶ arrows, or clicking the value to type it directly.
Clicking the LoRA name opens a searchable dropdown.

Actions

| Button | Action | | ---------- | ------------------------------------------------------- | | ➕ New | Creates a new style file from a blank template | | 💾 Save | Saves current widget values to the selected style file | | 📋 Clone | Saves current values to a new file | | 🗑️ Delete | Deletes the selected style file (requires confirmation) |

Adding a new style file or refreshing an existing one takes effect immediately without restarting ComfyUI. Adding new files to the styles/ folder externally requires a page refresh (F5) to appear in the dropdown.

CharacterLoader

CharacterLoader_v1

Inputs

character_file (combo) — dropdown listing all .json files found recursively under the src/data/characters/ folder.
eyes (str) — eye description tags.
hair_type (str) — hair type tags.
hair_style (str) — hair style tags.
makeup (str) — makeup tags.
jewelry (str) — jewelry tags.
body_type (str) — body type tags.

Outputs

eyes (str)
hair_type (str)
hair_style (str)
makeup (str)
jewelry (str)
body_type (str)

Usage

Same file system as StyleLoader, JSON files stored under src/data/characters/. Subfolders are supported and reflected in the dropdown after a page refresh.

{
  "eyes": "blue eyes, detailed eyes",
  "hair_type": "long hair, black hair",
  "hair_style": "ponytail, hair ribbon",
  "makeup": "eyeliner",
  "jewelry": "gold earrings, necklace",
  "body_type": "slim, tall"
}

Each field is output as a plain string, ready to be plugged into any other string input.

Actions

| Button | Action | | ---------- | ----------------------------------------------------------- | | ➕ New | Creates a new character file from a blank template | | 💾 Save | Saves current field values to the selected character file | | 📋 Clone | Saves current values to a new file | | 🗑️ Delete | Deletes the selected character file (requires confirmation) |

BreakEncoder

BreakEncoder_v1

Inputs

clip (CLIP) — CLIP encoder, typically from a checkpoint loader.
break_keyword (str), default BREAK — keyword used to split the prompt into blocks. Case-sensitive.
text (str) — multiline prompt with break blocks.
cleanup (bool), default True — applies prompt cleanup on each block before encoding.

Outputs

conditioning (CONDITIONING) — concatenated conditioning from all blocks.
text (str) — cleaned blocks joined by \n---\n, reflecting exactly what was encoded.

Usage

Write a prompt with break blocks separated by the keyword:

masterpiece, best quality,
BREAK
character, solo, long hair,
BREAK
white background, simple background

Each block is encoded separately then concatenated, equivalent to ComfyUI's ConditioningConcat chained across all blocks. This matches the behavior of A1111/Yodayo's BREAK keyword.

The text output reflects exactly what was encoded after cleanup, useful for debugging:

masterpiece, best quality
---
character, solo, long hair
---
white background, simple background

Cleanup

When enabled, applies the same cleanup rules as PromptPresetSelector on each block individually. Blocks that become empty after cleanup are silently filtered out.

Validation

The node stops the workflow on error if:

clip input is None
text is empty or whitespace only
break_keyword is empty or whitespace only
No valid blocks remain after splitting and cleanup

Web

contentEditable editor

Some text widgets are replaced by a contentEditable div that provides syntax highlighting. Supported widgets and nodes:

text widget in PromptPresetSelector
options widget in PromptOptionPicker
template widget in PromptLayoutFiller

Colors update in real time while editing or changing other widget's values and pasting always inserts plain text, stripping any HTML formatting.

PromptPresetSelector:

Orange: tags and separators
Gray: inactive presets
Blue: $N references on the active preset

PromptOptionPicker:

Orange: @combine / --- / @end tags
Cyan: labels
Gray: $: separators

PromptLayoutFiller:

Placeholders are highlighted in the editor:

Green: slot is connected
Red: slot is referenced but not connected

Keyboard shortcuts

| Shortcut | Action | | ----------- | ---------------------------------------------------------------------------------------------------------- | | Ctrl+Up | Increase weight of selected text or word under caret by 0.1. Ex: tag → (tag:1.1) | | Ctrl+Down | Decrease weight of selected text or word under caret by 0.1. Ex: (tag:1.1) → tag when reaching 1.0 |

The selection is preserved after each keypress, consistent with ComfyUI native behavior.

Tests

This project uses pytest and pytest-aiohtttp for its tests, and uv for the versioning. After downloading uv, just run:

uv sync --dev
.venv\Scripts\activate.bat    # Windows
source .venv/bin/activate     # Linux/MacOS

and you should have the dependencies ready.

Then, to run the tests:

uv run pytest

Limitations

Currently, the relative imports in the root __init__.py file are preventing pytest from running. The solution is to change all the imports from .src.xxx to src.xxx for the tests, and then revert them for the ComfyUI execution.

TODO

PromptPresetSelector

Add a field to easily concatenate text at the end of the output
Wildcard mode for presets (randomly selected, index to -1 or -2?)
Original preset_names unchanged if changed from a PromptRouter (even though the value is still the correct one)

PromptRouter

Add support for more sub-routing levels?
Support special node integration beyond one sub-router level?

StyleLoader

LoRA toggle on/off per row
Separate model/clip weight per LoRA on UI
Better LoRA row style

CharacterLoader

Preview image for characters

BreakEncoder

Colored editor highlighting BREAK keywords