ComfyUI Expert Wildcard Prompt

Expert Text Prompt for ComfyUI

A powerful ComfyUI custom node featuring AST-based prompt parsing, advanced wildcards, probabilistic weighting, skip chances, prompt grouping, dual positive/negative outputs, inline negative extraction (-), $negative placeholder injection, and inline Mute/Solo controls.

---

Features

• Advanced Wildcards: {option1 | option2 | option3} with full support for nested wildcards.
  • Equal Chances: {red | blue | green} assigns an equal chance to each option.
  • Weighted Chances: {70% blue | 30% green} sets explicit relative probabilities.
• Skip Chance (Optional Blocks): {20%? optional sunglasses, leather jacket, red hat} (20% chance to skip the ENTIRE block completely).
• Prompt Grouping ([GRP:NAME]): Structure large prompts. Groups can be redefined/overwritten ([GRP:NAME]), appended to ([+GRP:NAME]), killed (-[GRP:NAME]), muted (//[GRP:NAME]) or set to solo (![GRP:NAME]).
• Inline Mute (//): Disable specific tags or entire prompt groups without deleting them. Multiple // tags can be used simultaneously.
• Inline Solo (!): Isolate specific tags or groups, ignoring all non-solo elements. Multiple ! tags can be used to keep a specific set of tags active.
• Quadruple Outputs (positive, negative, combinations & debug_info): Generates positive and negative strings, outputs live combination counts (INT), and provides a debug_info (STRING) pin displaying visual syntax error diagnostics with ^ pointers.
• Inline Negative Extraction & Subtractive Filtering (-): Prefix any tag or wildcard option with - (e.g., -sunglasses, -umbrella). This routes the tag into the negative output AND acts as a smart filter, actively removing any positive tags that contain this exact word!
• $negative Placeholder Injection: Insert $negative into your negative prompt field to specify the exact location where extracted - tags are injected.
• Flexible Negative Modes:
  • auto (use $negative): Injects extracted tags into the $negative placeholder if present, otherwise prepends them.
  • prepend: Places extracted negative tags at the very start.
  • append: Appends extracted negative tags at the very end.
  • replace: Overwrites the negative prompt field completely.
• Number Range Wildcards ({MIN-MAX:STEP}): {18-50} randomly selects a number in range. Supports optional steps like {0-10:2} (chooses from 0, 2, 4, 6, 8, 10).
• SDXL Weights & LoRAs: Native preservation of (tag:1.2) weights and <lora:name:1.0> tags.
• Deterministic Seed: Reproducible wildcard resolution based on input seed.

---

Installation

Option 1: Via Comfy Registry (CLI)

comfy node install comfyui-expert-wildcard-prompt

Option 2: Via ComfyUI Nodes Manager (UI)

Search for ComfyUI Expert Wildcard Prompt in the ComfyUI Nodes Manager / Extension Installer and click Install.

Option 3: Manual Git Clone

Navigate to your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/v3rm1ll1on/ComfyUI-Expert-Wildcard-Prompt.git

Restart ComfyUI afterward.

---

Finding the Node in ComfyUI

• Node Name: Expert Text Prompt (Wildcards & AST)
• Category: prompt/expert

---

Full Syntax Reference (Cheat Sheet)

| Feature | Syntax Example | Description | | :--- | :--- | :--- | | Basic Wildcard | {red \| blue \| green} | Pick one option randomly with equal chance. | | Weighted Wildcard | {70% blue \| 20% green \| 10% red} | Assign custom percentage probabilities per option. | | Skip Chance (Optional) | {20%? sunglasses} | 20% chance to omit tag completely (80% chance to include). | | Number Range | {18-50} | Random integer between min (18) and max (50). | | Number Range with Step | {1980-2020:5} | Pick random integer with step (1980, 1985, ..., 2020). | | Nested Wildcards | {60% female, {70% armor \| 30% suit} \| 40% male} | Multi-level wildcard nesting with combined probabilities. | | Inline Mute Tag | // leather jacket | Temporarily deactivates tag without deleting it. | | Inline Solo Tag | ! red dress | Isolates marked tags. Non-solo tags are ignored. | | Prompt Group | [GRP:NAME], tag1, tag2 | Groups tags into a structured, manageable block. Redefining it later overwrites the group. | | Append to Group | [+GRP:NAME], tag3 | Appends tags to an existing group without overwriting it. | | Kill / Mute Group | -[GRP:NAME] or //[GRP:NAME] | Completely clears the group from positive output (Mute) and optionally routes to negative (Kill). | | Inline Negative Extraction | -fish | Routes tag to negative output AND actively removes any positive tags containing the exact word "fish" (e.g. removes "green fish" but keeps "starfish"). | | Negative Placeholder | $negative | Specifies exact insertion point for - tags in negative_prompt. | | SDXL Weights & LoRAs | (masterpiece:1.2), <lora:name:1.0> | Preserves weight syntax and LoRA tags natively. |

---

Usage Examples & Templates

We provide a dedicated examples/ directory with detailed Markdown templates ordered from beginner to masterclass:

| Example File | Topic & Level | Description | | :--- | :--- | :--- | | 01_basic_wildcard.md | Beginner | Step-by-step introduction from single wildcards to nested wildcard structures. | | 02_intermediate_groups_and_weights.md | Intermediate | Percentage weighting (70%), skip chances (20%?), and introducing groups ([GRP:]). | | 03_advanced_pony_v6_template.md | Advanced | Specialized templates for Pony Diffusion V6 with score groups & anime styles. | | 04_expert_photorealistic_dual_prompt.md | Expert | Photorealistic templates combining - negative extraction and $negative. | | 05_mute_and_solo_indepth.md | In-Depth Guide | Tag-level and group-level Mute (//) & Solo (!) isolation workflows. | | 06_nested_wildcards_indepth.md | In-Depth Guide | 2-Level and 3-Level wildcard nesting in plain text and group contexts. | | 07_masterclass_character_generator.md | Masterclass | Full production template for multi-ethnic, hyper-realistic character generation. |

---

Quick Syntax Summary

1. Basic Wildcards

Randomly choose between simple options with equal probability:

a photo of a {cat | dog | fox} sitting on a bench

2. Weighted Chances & Smart Weight Balancing

Explicitly set selection probabilities or mix weighted and unweighted options:

a photo of a girl with {70% blue eyes | 20% green eyes | 10% brown eyes}

How Smart Weight Balancing Works:

You don't need to manually calculate probabilities so everything sums to 100%. The node automatically balances weights intelligently:

• 1. Automatic Leftover Distribution (Partial Weights) If you only weight some options, the remaining percentage is split equally among unweighted options:

  {50% red | 30% green | black | white}
  

  • red (explicit): 50%
  • green (explicit): 30%
  • Explicit sum = 80%. Remaining budget = 20%.
  • black & white split leftover 20% equally $\rightarrow$ 10% each.
  • Final Odds: red: 50% | green: 30% | black: 10% | white: 10%.

• 2. Over-Commitment Graceful Fallback (Sum > 100%) If your explicit weights exceed 100%:

  {50% red | 30% green | black | 50% white}
  

  • Unweighted tags (black) automatically receive a baseline fair share ($100% / N_{\text{total}} = 25%$).
  • Relative proportions are preserved smoothly without ignoring any option.

• 3. Zero-Sum Safety Guard If all options are set to 0% ({0% red | 0% green}), the node automatically falls back to equal distribution (50%/50%) instead of crashing.

[!WARNING] Weights (XX%) vs. Skip Chance (XX%?) Do not confuse probability weights with skip chances! If you want an option to appear rarely (e.g. 5%), use a normal weight: {blonde | 5% brunette | black | red}. The ? syntax (e.g. {80%? ...}) is EXCLUSIVELY used at the very start of a wildcard to skip the entire block, yielding an empty string. If you place a ? inside an option list, it will just be parsed as a normal weight.

3. Skip Chance (Optional Blocks)

Add optional elements with a percentage chance of skipping:

portrait of a woman, {20%? glowing neon face tattoos, multiple piercings, dark makeup}

(20% chance to omit all of these tags entirely, 80% chance to include the whole block).

4. Number Range Wildcards ({MIN-MAX:STEP})

Generate random numeric values across a range with optional step increments:

a photo of a {18-50} yo woman born in {1980-2020:5}

(Chooses an age between 18 and 50, and a year from [1980, 1985, 1990, ..., 2020]).

5. Mute & Solo Controls (// & !)

Temporarily disable or isolate tags without editing your prompt text:

• Mute (//): Deactivates specific tags (supports multiple // in one prompt):

  masterpiece, // ruined background, leather jacket, // blurry lines
  

• Solo (!): Isolates only marked tags (supports multiple ! tags):

  ! red dress, blue shoes, ! golden necklace
  

  (Resulting output: red dress, golden necklace)

5. Nested Wildcards & Probabilities

Combine multiple levels of wildcards for complex variation:

a {60% female warrior with {70% knight armor | 30% cyber suit} | 40% cyberpunk rogue}

6. Dynamic Prompt Grouping ([GRP:NAME], [+GRP:NAME], -[GRP:NAME])

Structure large prompts into organized blocks that can be dynamically manipulated anywhere, even inside wildcards!

• Overwrite ([GRP:NAME]): Clears any previous tags for this group and starts fresh.
• Append ([+GRP:NAME]): Adds tags to the existing group without deleting old ones.
• Kill / Mute (-[GRP:NAME] / //[GRP:NAME]): Completely removes the entire group from the positive output.

[GRP:EXPRESSION], happy, smiling

[GRP:ACTION],
{
  [GRP:POSE1], standing around, [+GRP:EXPRESSION], laughing
  |
  [GRP:POSE2], running away, [GRP:EXPRESSION], exhausted, sweaty
  |
  [GRP:POSE3], jumping, -[GRP:EXPRESSION]
}

(If POSE1 is rolled, EXPRESSION becomes happy, smiling, laughing. If POSE2 is rolled, EXPRESSION is overwritten to just exhausted, sweaty. If POSE3 is rolled, EXPRESSION is completely deleted!)

7. Expert Dual-Prompting & Subtractive Filtering (- & $negative)

Combine positive wildcards, automatic negative extraction, and custom negative templates in one go. The - tag also acts as an active subtractive filter for your positive prompt using whole-word matching.

positive_prompt:

[GRP:CHARACTER], portrait of a young woman, {70% sunny day, -sunglasses | 30% rainy day, -umbrella}, leather jacket, green umbrella

negative_prompt:

(3d render, cgi, plastic skin:1.3), $negative, (deformed hands:1.2), blurry

Output Results:

• If sunny day is selected by the seed:
  • positive: portrait of a young woman, sunny day, leather jacket, green umbrella
  • negative: (3d render, cgi, plastic skin:1.3), sunglasses, (deformed hands:1.2), blurry
• If rainy day is selected by the seed:
  • positive: portrait of a young woman, rainy day, leather jacket (Notice how green umbrella is actively deleted from the positive prompt because of -umbrella!)
  • negative: (3d render, cgi, plastic skin:1.3), umbrella, (deformed hands:1.2), blurry

---

Node Parameters & Outputs

• Inputs:
  • positive_prompt: Main prompt text supporting wildcards, Mute/Solo, weights, and - negative tags.
  • negative_prompt: Secondary prompt field supporting the $negative placeholder.
  • negative_mode: Options: auto (use $negative), prepend, append, replace.
  • seed: Random seed for reproducible wildcard selections.
• Outputs:
  • positive: Connects to CLIP Text Encode (Positive).
  • negative: Connects to CLIP Text Encode (Negative).

---

License

MIT