ComfyUI-Prompt_util_pack

✅ README.md — ComfyUI-prompt-utils

🧠 ComfyUI Prompt Utils

A modular toolkit for advanced prompt engineering in ComfyUI — designed for dataset creation, LoRA training, batch generation, and controlled variation.

✅ Extract & inject variables (seed=123, model=epicrealism)
✅ Manage multi-prompt batches (001_smile\n...\n---\n002_serious)
✅ Hide & swap variations with [[smile;laugh;neutral]]
✅ Fully composable — nodes work independently or chained

💡 No external dependencies — pure Python, works with any model (SDXL, Flux, Wan, LoRA, etc.).

---

📦 Installation

In your ComfyUI/custom_nodes folder:

git clone https://github.com/fmartinellidev/ComfyUI-prompt-utils.git

Or download as ZIP and extract into ComfyUI/custom_nodes/ComfyUI-prompt-utils.

Restart ComfyUI.

---

🧩 Included Nodes

| Node | Purpose | |------|---------| | Prompt Snippet Extractor | Split multi-prompt text into indexed snippets | | Prompt Variable Extractor | Parse key='value' or key="value" from text | | Prompt Variable Substitutor | Replace {key} placeholders with values | | Prompt Hidden Processor | Swap [[option1;option2]] with a selected variation |

All nodes are independent — use one or chain them all!

---

🔍 Node Details

1. Prompt Snippet Extractor

Splits a multi-prompt document (e.g. for dataset generation) into indexed snippets.

🔤 Inputs

| Input | Type | Required | Default | Description | |-------|------|----------|---------|-------------| | prompt_list | STRING | ✅ | (example) | Multi-snippet text (delimited by --- or custom) | | snippet_index | INT | ✅ | 0 | Which snippet to extract (0-based) | | split_char | STRING | ✅ | "---" | Delimiter between snippets | | first_word_is_filename | BOOLEAN | ✅ | True | Treat first word as filename source | | ignore_start_number_label | BOOLEAN | ✅ | True | Strip leading numbers (e.g. 001_ → "") |

📤 Outputs

| Output | Type | Description | |--------|------|-------------| | prompt | STRING | Full snippet (e.g. "001_smile_front\nA gentle smile...") | | prompt_text | STRING | Snippet without first word (e.g. "A gentle smile...") | | filename_to_label | STRING | Cleaned label (e.g. "smile front") | | filename | STRING | First word (e.g. "001_smile_front") |

💡 Example

Input (prompt_list):

001_smile_front  
A gentle smile, facing camera  
---  
002_laugh_dynamic  
Laughing joyfully, head tilted

With snippet_index=0, first_word_is_filename=True, ignore_start_number_label=True:

| Output | Value | |--------|-------| | prompt | "001_smile_front\nA gentle smile, facing camera" | | prompt_text | "A gentle smile, facing camera" | | filename_to_label | "smile front" | | filename | "001_smile_front" |

---

2. Prompt Variable Extractor

Extracts key='value' or key="value" pairs from text (e.g. metadata, config).

🔤 Inputs

| Input | Type | Required | Default | Description | |-------|------|----------|---------|-------------| | text | STRING | ✅ | "" | Input text containing variables | | prefix | STRING | ✅ | "#" | Prefix before keys (e.g. #seed=123) | | delimiter_char | STRING | ✅ | "=" | Char between key and value |

📤 Outputs

| Output | Type | Description | |--------|------|-------------| | output_text | STRING | Text with variables removed | | variables_json | STRING | JSON of extracted variables (e.g. {"seed": "123"}) |

💡 Example

Input (text):
#seed=1234 #model="epicrealismxl" A portrait, studio lighting

Outputs:

• output_text: "A portrait, studio lighting"
• variables_json: {"seed": "1234", "model": "epicrealismxl"}

---

3. Prompt Variable Substitutor

Replaces {key} placeholders in a template using a JSON variable map.

🔤 Inputs

| Input | Type | Required | Description | |-------|------|----------|-------------| | prompt_template | STRING | ✅ | Template with {key} placeholders | | variables_json | STRING | ✅ | JSON object (e.g. {"seed": "123"}) |

📤 Outputs

| Output | Type | Description | |--------|------|-------------| | prompt | STRING | Prompt with placeholders replaced |

💡 Example

Inputs:

• prompt_template: "A {expression} portrait, {lighting} lighting"
• variables_json: {"expression": "smiling", "lighting": "soft"}

Output:

• prompt: "A smiling portrait, soft lighting"

---

4. Hidden Prompt Processor

Swaps hidden variation blocks like [[smile;laugh;neutral]] with a selected option.

🔤 Inputs

| Input | Type | Required | Default | Description | |-------|------|----------|---------|-------------| | prompt | STRING | ✅ | (example) | Prompt with hidden blocks | | input_index_split | INT | ✅ | 0 | Which variation to use (0-based) | | delimiter_pair | STRING | ✅ | "[[ ]]" | Block style: [[ ]] or ## ## |

📤 Outputs

| Output | Type | Description | |--------|------|-------------| | input_prompt | STRING | Prompt with hidden block replaced | | clean_prompt | STRING | Prompt with all hidden blocks removed |

💡 Example

Input (prompt):
A portrait of [[smile;laugh;neutral]], {lighting} lighting

With input_index_split=1, delimiter_pair="[[ ]]":

| Output | Value | |--------|-------| | input_prompt | "A portrait of laugh, {lighting} lighting" | | clean_prompt | "A portrait of , {lighting} lighting" |

🔗 Pro tip: Chain with Prompt Variable Substitutor to replace {lighting} next!

---

🧩 Recommended Workflow

graph LR
    A[Prompt Snippet Extractor] -->|prompt_text| B[Hidden Prompt Processor]
    B -->|input_prompt| C[Prompt Variable Substitutor]
    C --> D[Stable Diffusion / WaveSpeed / etc.]
    A -->|filename| E[SaveImage]

Perfect for:

• 📁 LoRA dataset generation (200+ expressions)
• 🎭 Controlled facial variation (no random sampling)
• 📊 Metadata-rich outputs (embed seed/model in filename)

---

🤝 Contributions

PRs welcome! For major changes, please open an issue first.

---

Made with ❤️ by fmartinellidev
Inspired by advanced prompt engineering needs in AI character creation.