Wizdroid Character Nodes for ComfyUI

Version: 2025.11.01

A streamlined collection of custom nodes for ComfyUI that provide AI-powered prompt generation tools using Ollama. These nodes focus on detailed character creation, cultural storytelling, image analysis, and remix workflows.

Features

Character Prompt Builder

• Purpose: Generates detailed character prompts for image generation with extensive customization options
• Attributes: Character name, gender, age group, body type, hair color/style, eye color, facial expressions, poses, makeup, fashion styles, upcycled fashion materials, and backgrounds
• Integration: Ollama LLM integration for intelligent prompt crafting
• Controls: LLM token-limit dropdown (128–4096 tokens in 128-token steps; applies only to the Ollama model) plus workflow seed control
• Camera lens: camera_lens uses concise focal-length or lens-type names (24mm, 50mm, 85mm, 70-200mm, fisheye, anamorphic, medium format, etc.), without attached descriptions.
• Color palettes: color_palette contains practical photoshoot palettes (black and white, warm neutrals, muted earth tones, teal & orange, high-contrast, studio white) rather than creative/abstract names.
• Pose controls: pose_content_rating toggle (SFW / NSFW / Mixed) paired with SFW and NSFW pose catalogs for safer randomization
• Output: Positive prompt, negative prompt, and a preview string (matches the positive prompt)

Character Edit Node

• Purpose: Takes an existing prompt and nudges it toward refined directions (fashion, lighting, mood, etc.)
• Tools: Section-by-section emphasis controls with Ollama-powered rewrites
• Controls: Shares the same LLM token dropdown (128–4096 tokens) to keep Ollama generations on budget
• Pose controls: Select whether edits should draw from the SFW or NSFW pose pool before randomization
• Output: Revamped prompt plus diff-style summary for reference

Prompt Combiner Node

• Purpose: Merge multiple prompt fragments into a cohesive description
• Features: Adjustable weights, connective phrasing, and automated deduplication
• Controls: Uses the same Ollama token dropdown to enforce consistent LLM verbosity across merged prompts
• Output: Blended prompt ready for downstream nodes

Photo Aspect Extractor

• Purpose: Analyzes images using vision models to extract specific aspects
• Modes: Clothes, pose, style, background, expression, lighting, hair, makeup, accessories, camera settings, composition, color palette
• Integration: Ollama vision models (LLaVA, Florence, etc.) for image analysis
• Output: Extracted aspect descriptions for prompt engineering

LoRA Dataset Export ✨

• Purpose: Creates Kohya-compatible datasets (images/, captions.jsonl, metadata) from local image folders or comma-separated paths
• Prompting: Offers manual captions, character-builder prompts, or Ollama vision analysis with the exact same model dropdown the Photo Aspect Extractor exposes (auto-discovers vision models via Ollama’s /api/tags)
• Image Prep: Optional resize toggle (none, longest_1024, longest_512) to keep inputs consistent before training
• Output: Returns dataset folder path plus captions file for downstream validation/training nodes

Tip: Every LLM-driving node now exposes a token_limit_override dropdown (128–4096 tokens, default 128). It only governs Ollama’s response length, not your downstream image generator, so you can standardize prompt verbosity without touching prompt_styles.json.

Installation

1. Clone this repository into your ComfyUI custom_nodes directory:

   cd /path/to/ComfyUI/custom_nodes
   git clone https://github.com/yourusername/wizdroid-character.git
   

2. Install the required dependencies:

   pip install -r requirements.txt
   

3. Ensure Ollama is running with appropriate models:

   • For text generation: Install models like llama2, mistral, or codellama
   • For vision analysis: Install vision models like llava, bakllava, or florence

Configuration

The nodes use JSON configuration files in the data/ directory:

• character_options.json: Character attributes, upcycled materials, and styling options
• countries.json: Specific countries for fashion inspiration
• regions.json: Broader regional areas for fashion inspiration
• prompt_styles.json: Output format styles for different AI models
  • Includes presets for Flux, SD/SDXL, SDXL-Turbo, Juggernaut, RealVis, HiDream, and Qwen editing formats Note: fantasy_options.json and the Fantasy Scene node have been removed from this package. If you previously relied on fantasy scene generation, please migrate any workflows to the Character Prompt Builder or add a custom node. Former glamour_options live inside the pose_style.sfw / pose_style.nsfw blocks in character_options.json, and the unused followup_questions.json catalog has been retired.

Usage

After installation, the nodes will appear in ComfyUI under the "Wizdroid/character" category.

Common Workflow

1. Use Character Prompt Builder to generate richly detailed prompts
2. Use Photo Aspect Extractor (or your preferred external pose tool) to analyze reference images
3. Combine extracted elements with generated prompts for refined results via Prompt Combiner
4. Iterate with Character Edit Node to explore alternate directions

Requirements

• ComfyUI
• Ollama server running locally
• Python packages: requests, torch, PIL (Pillow)
• For vision features: Ollama vision models

Contributing

Feel free to contribute by:

• Adding more cultural outfits and countries
• Expanding attribute options in JSON files
• Improving prompt generation logic
• Adding new node types
• Enhancing vision model integrations

License

[Add your license here]