Boyonodes (Vision Nodes Disabled due to the dependencies causing conflicts, will resolve soon)

Essential nodes for semantic image editing workflows with Kontext, Qwen Image Edit, and HiDream E1.1 models. Streamlines iterative editing, dataset creation, and batch processing for next-generation semantic editing pipelines. Now with Lora loader for Wan2.2 which automatically selects pairs and any trigger words for you.

Semantic Image Editing Workflow Nodes

Boyo Image Grab

Automatically monitors directories and grabs the most recent image file - perfect for iterative semantic editing workflows where you want to chain multiple edits together without manual file management.

Key Features:

• Auto-detects newest image in specified directory
• Enables seamless iterative editing chains
• Perfect for progressive semantic modifications
• Supports all major image formats
• Real-time directory monitoring

Inputs:

• directory_path: Full path to monitor (e.g., "C:\MyImages")
• file_extensions: Comma-separated formats (default: "jpg,jpeg,png,bmp,tiff,webp")
• auto_refresh: Enable automatic monitoring (default: True)
• refresh_trigger: Manual refresh control (default: 0)

Outputs:

• image: Ready-to-use image tensor
• filename: Just the filename
• full_path: Complete file path
• timestamp: Modification timestamp

Usage: Perfect for workflows like: Original â†' "add a hat" â†' "make it red" â†' "change lighting" where each step automatically feeds into the next.

Boyo Paired Image Saver

Saves original and edited image pairs with organised sequential naming - essential for creating datasets from semantic editing workflows.

Boyo Paired Video Saver

Same as above but for video.

Key Features:

• Maintains original/edited image relationships
• Sequential batch numbering
• Organised dataset creation
• ControlNet data format compatible
• Custom directory placement (inside or outside ComfyUI)

Inputs:

• original_image: Source image
• controlnet_image: Edited/processed image
• folder_name: Target directory
• filename_prefix: File naming prefix

Output Format:

generated001_original.png
generated001_controlnet.png
generated002_original.png
generated002_controlnet.png

Boyo Incontext Saver

Specialised saver for semantic editing model outputs with proper dataset formatting.

Key Features:

• Handles source and diffusion outputs
• Dataset-ready file organisation
• Batch processing support
• Maintains editing relationships

Inputs:

• source_image: Original input
• diffusion_output: Semantic edit result
• folder_name: Output directory

Boyo Universal Image + Prompt Saver

Saves clean images with accompanying prompt files - perfect for publication and documentation.

Key Features:

• Strips metadata for clean publication
• Saves actual prompts (crucial for wildcard/enhanced workflows)
• Creates organised image/text pairs
• Protects workflow secrets whilst maintaining documentation

Perfect for:

• Publishing images without exposing ComfyUI metadata
• Documenting wildcard-generated prompts you never see
• Creating training datasets with prompt annotations
• Maintaining reproducible results

LoRA Management System

Revolutionary paired LoRA management with intelligent prompt handling - perfect for complex workflows requiring multiple LoRA types (utility, character, style) with sophisticated prompt strategies.

Boyo LoRA JSON Builder

Create and save LoRA configuration files with flexible prompt and file assignments.

Key Features:

• Supports paired LoRAs (high/low noise variants)
• Flexible prompt management (multiple prompts per config)
• Handles utility LoRAs (no prompts needed)
• Graceful handling of single LoRAs or config-only entries
• Auto-saves to organised directory structure

Inputs:

• name: Configuration name (becomes filename)
• prompts: Multi-line text (one prompt per line, optional)
• high_noise_lora: LoRA file selector (optional)
• low_noise_lora: LoRA file selector (optional)
• save_config: Boolean toggle to save JSON

Perfect for:

• Lightning/reward LoRAs (utility configs with no prompts)
• Character LoRAs with example prompts
• Style LoRAs with trigger words
• Any LoRA requiring organised management

Boyo LoRA Paired Loader

Load multiple LoRA configurations simultaneously with intelligent prompt handling strategies.

Key Features:

• 3 simultaneous config slots for layered effects (utility + character + style)
• Prompt strategies per config: Mute, Concatenate, Merge
• Prompt modes: First Only, Cycle Through, Random (seed-based)
• Smart error handling - continues with missing configs
• Combined prompt output for easy workflow integration

Inputs:

• prompt_mode: Global prompt selection behaviour
• lora_config_1/2/3: Configuration file selectors
• prompt_strategy_1/2/3: Individual prompt handling per config
• seed: For consistent random prompt selection

Outputs:

• 6 LoRA paths: high_noise_path_1/2/3, low_noise_path_1/2/3 (connect directly to LoRA loaders)
• 4 prompt strings: individual prompts + combined output

Workflow Integration: Outputs connect directly to any standard LoRA loader. No primitives needed - clean, direct connections.

Boyo LoRA Config Inspector

Preview LoRA configuration contents before loading for informed decision-making.

Key Features:

• Smart analysis: Detects utility, paired, or single LoRA configs
• File status checking: Verifies LoRA files actually exist
• Usage recommendations: Suggests optimal prompt strategies
• Three output formats: formatted summary, raw JSON, status line
• Real-time inspection: See exactly what's in each config

Perfect for:

• "What was in that config again?" scenarios
• Debugging missing LoRA files
• Planning complex multi-LoRA workflows
• Understanding prompt availability before loading

Sample Output:

ðŸ"‹ LoRA Configuration: Character_Cyborg
🎯 LoRA FILES:
  ðŸ"ˆ High Noise: ✅ cyborg_char_v2.safetensors
  ðŸ"‰ Low Noise: ✅ cyborg_char_v2_low.safetensors
  🎭 Type: PAIRED LoRA (different high/low files)

ðŸ'¬ PROMPTS (3 total):
  1. cyborg woman, metallic skin, glowing eyes
  2. android female, chrome details, futuristic
  3. robotic humanoid, synthetic appearance

ðŸ'¡ USAGE RECOMMENDATIONS:
  • Use 'Cycle Through' for variety
  • Use 'Random' for experimentation

AI Storyboard Generation System

Automated storyboard prompt generation using ollama models for consistent multi-scene video workflows. Perfect for Next Scene LoRA workflows and traveling prompt video generation.

Boyo Storyboard Prompt

Intelligent prompt generator that creates structured storyboard sequences using local ollama models.

Key Features:

• Model-agnostic trigger words - works with any LoRA or video model
• Two generation modes: Standard 6-scene storyboards or traveling prompt sequences
• Automatic JSON formatting for clean workflow integration
• Consistent character/style maintenance across all scenes
• Optimised for abliterated coder models (Qwen 30B A3B Coder recommended)

Inputs:

• story_concept: Multi-line story description
• main_character: Detailed character description
• style_setting: Art style, mood, technical specifications
• image_trigger_word: Model-specific trigger (e.g., "Next Scene:", "Character:")
• video_trigger_word: Video model trigger (e.g., "Motion:", "Camera:")
• system_prompt_type: Choose generation mode
• traveling_prompt_count: Number of sub-prompts per scene (System Prompt 2 only)
• additional_details: Extra instructions (use for verbosity control)

System Prompt 1 (6 Scenes):

• Generates 6 detailed image prompts for storyboard
• Generates 6 corresponding video prompts with camera movements
• Perfect for Next Scene LoRA workflows requiring scene-by-scene consistency

System Prompt 2 (Traveling Prompts):

• Generates 6 detailed image prompts for storyboard foundation
• Generates 6 video scenes, each containing multiple traveling sub-prompts
• Each sub-prompt ≈ 5 seconds of video content
• Example: 3 traveling prompts = 6 scenes × 3 prompts = 18 total video prompts

Outputs:

• Formatted prompt string ready for ollama integration

Recommended Models:

• Best quality: Qwen 30B A3B Coder Abliterated
• Best speed/efficiency: Nemo 7B with verbosity nudging
• Avoid: Google models (Gemma), Meta coding variants, thinking models

Boyo Storyboard Output

Parses ollama JSON responses into separate prompt outputs for direct workflow integration.

Key Features:

• Robust JSON parsing with automatic markdown cleanup
• 12 separate outputs for System Prompt 1 (6 image + 6 video)
• Multi-line support for traveling prompts in System Prompt 2
• Graceful error handling with descriptive error messages
• Direct workflow connection - outputs connect straight to prompt inputs

Input:

• json_input: Raw JSON response from ollama

Outputs:

• image_scene1 through image_scene6: Individual image prompts
• video_scene1 through video_scene6: Individual video prompts (single or multi-line)

JSON Structure Handled:

{
  "imagePrompts": {
    "scene1": "Detailed image prompt...",
    "scene2": "Detailed image prompt..."
  },
  "videoPrompts": {
    "scene1": "Video prompt for System Prompt 1 OR multi-line traveling prompts for System Prompt 2",
    "scene2": "Video prompt for System Prompt 1 OR multi-line traveling prompts for System Prompt 2"
  }
}

Integration Workflow:

1. Boyo Storyboard Prompt → ollama Generate → Boyo Storyboard Output
2. Connect 12 outputs directly to image/video generation workflows
3. Supports both standard scene generation and extended traveling prompt sequences

Workflow Enhancement Nodes

Boyo Empty Latent

Smart aspect ratio calculator - input width, select ratio, automatically calculates height.

Features:

• Model-aware resolution calculation
• Common aspect ratio presets
• Prevents invalid dimensions

Load Image List

Batch image processor for mass workflows.

Features:

• Mass resize, pad, or crop operations
• Perfect for upscaler pipelines
• Handles hundreds of images automatically
• Set-and-forget batch processing

Utility Nodes

BoyoVision Node

Qwen2.5VL vision node with abliterated version compatibility. The abliterated model is uncensored and therefor does not refuse tasks given to it. Far superior and accurate as a result

Boyo VAE Decode Nodes

Stealth NSFW filtering for controlled environments.

Standard VAE Decode:

• Hidden NSFW detection
• Disguisable as normal VAE node
• Perfect for educational/workplace deployment

Tiled VAE Decode:

• Memory-efficient processing for large images
• Seamless tile blending
• VRAM-friendly for high-resolution workflows
• Same stealth NSFW protection

Mandelbrot Video Generator

Fractal art generation for creative projects.

Features:

• Random Mandelbrot and Julia sets
• Video or frame sequence output
• Multiple resolution options
• Grayscale mode available

Installation

Quick Install:

git clone https://github.com/DragonDiffusionbyBoyo/Boyonodes.git
cp -r Boyonodes /path/to/ComfyUI/custom_nodes/

Most nodes require only ComfyUI itself.

Mandelbrot Video Node requires additional dependencies:

pip install numpy==1.26 matplotlib pillow tqdm torch

FFmpeg (for video output):

• Windows: Download from ffmpeg.org or choco install ffmpeg
• macOS: brew install ffmpeg
• Linux: sudo apt-get install ffmpeg

Restart ComfyUI after installation.

LoRA Management Workflow Examples

Basic Multi-LoRA Setup:

1. Create configs for utility (lightning), character, and style LoRAs
2. Load all three simultaneously in Paired Loader
3. Set utility to "Mute", character to "Concatenate", style to "Concatenate"
4. Get combined prompts and all LoRA paths in one node

Progressive LoRA Development:

1. Build character LoRA config with multiple example prompts
2. Use "Cycle Through" mode for testing different prompt variations
3. Inspector node helps verify which prompts are available
4. Iterate and refine prompt collections

Dataset Creation with LoRAs:

1. Set up configs for consistent LoRA + prompt combinations
2. Use "Random" mode with fixed seeds for controlled variation
3. Paired Image Saver captures results with LoRA-enhanced prompts
4. Creates organised datasets with known LoRA configurations

Semantic Editing Workflow Examples

Basic Iterative Editing:

1. Load initial image
2. Run through Kontext/Qwen/HiDream with prompt
3. Boyo Paired Image Saver stores original + edit
4. Boyo Image Grab automatically picks up the edit
5. Feed back for next iteration
6. Repeat for progressive modifications

Dataset Creation:

1. Use semantic editing models on source images
2. Boyo Paired Image Saver organises results
3. Creates properly formatted training datasets
4. Maintains source/target relationships

Publication Workflow:

1. Create final edited images
2. Boyo Universal Saver strips metadata
3. Saves clean images + prompt documentation
4. Ready for sharing without exposing workflows

Storyboard Generation Workflow Examples

Basic Storyboard Creation (System Prompt 1):

1. Configure story concept, character, and style in Boyo Storyboard Prompt
2. Set trigger words for your specific image/video models
3. Connect to ollama Generate node
4. Parse results with Boyo Storyboard Output
5. Connect 6 image outputs to Next Scene LoRA workflow
6. Connect 6 video outputs to video generation pipeline

Extended Video Sequences (System Prompt 2):

1. Set traveling_prompt_count (e.g., 6 for 30-second scenes)
2. Generate 6 storyboard images + 6 multi-line video prompt sequences
3. Each video scene contains multiple traveling prompts for extended sequences
4. Perfect for longer narrative videos with scene consistency

Character Consistency Workflow:

1. Use detailed character descriptions in main_character field
2. Include style/lighting preferences in style_setting
3. Leverage trigger words optimised for your LoRA models
4. Generate consistent character appearance across all 6 scenes

Troubleshooting

LoRA Management issues:

• Verify LoRA files exist in specified subdirectories
• Check JSON syntax in config files
• Use Inspector node to debug config problems
• Ensure LoRA paths use forward slashes

Boyo Image Grab issues:

• Verify directory path exists and is accessible
• Check file permissions
• Ensure auto_refresh is enabled
• Use forward slashes in Windows paths

Paired Saver issues:

• Confirm output directory exists
• Check file write permissions
• Verify input image formats

Storyboard Generation issues:

• Use recommended ollama models (Qwen 30B A3B Coder Abliterated)
• Add verbosity instructions in additional_details field
• Verify JSON structure with Boyo Storyboard Output error messages
• Check ollama connectivity and model availability

Performance tips:

• Large directories may slow Image Grab scanning
• Organise files into smaller subdirectories
• Use specific file extensions to reduce scanning
• LoRA configs are cached for faster loading
• Abliterated coder models perform best for storyboard generation

Contributing

Fork, branch (git checkout -b feature-name), commit, push, and open a pull request. Documentation for new features is appreciated.

License

MIT License - see LICENSE file for details.

Built by DragonDiffusionbyBoyo for the semantic editing revolution.