A collection of utility nodes designed to enhance your ComfyUI workflow with smart image loading, metadata management, widget extraction, Krita integration and more.

Category: TOO-Pack

📦 Installation

cd ComfyUI/custom_nodes
git clone https://github.com/tetsuoo-online/Comfyui-TOO-Pack

Requirements

ComfyUI (latest version recommended)
Python 3.8+
PyTorch
Pillow (PIL)
NumPy

🎯 Nodes Overview

| Node | Category | Description | |------|----------|-------------| | Simple Image Loader | TOO-Pack/image | Simple image loader with a dynamic preview, no need to start the workflow :) | | Smart Image Loader | TOO-Pack/image | Flexible image loader with multiple sources | | Smart Image Saver | TOO-Pack/image | Intelligent saver with flexible naming and metadata | | Smart Image Saver (Advanced) | TOO-Pack/image | Advanced saver with dynamic naming and A1111/Civitai metadata | | TOO Crop Image | TOO-Pack/image | Interactive cropping tool | | TOO Lora Grid | TOO-Pack/image | Special KSampler for making LoRA grids | | TOO Lora Grid Advanced | TOO-Pack/image | Advanced KSampler for making LoRA grids - Adapted from SamplerCustomAdvanced node | | TOO Image Grid From Batch with Alpha | TOO-Pack/image | Grid from batch that preserves the alpha channel | | TOO lmage Metadata | TOO-Pack/image | Metadata editing | | Krita Bridge | TOO-Pack/image | Auto-load images from Krita | | Extract Widget From Node | TOO-Pack/utils | Extract widget values from workflow nodes | | Collection Categorizer | TOO-Pack/utils | Categorize files with local LLM (Ollama) | | Time Converter | TOO-Pack/utils | Quick tool for converting seconds/min/hours |

📚 Node Documentation

Simple Image Loader 🖼

<details> <summary>Click to expand full documentation</summary>

An image loader that show instant preview with any path. You're no longer limited to ComfyUI/input folder. The idea was inspired by the Load Image (Path) node from comfyui-VideoHelperSuite, with a few more features.

Features

🖼️ Image input via path img_path will work with any path on your computer, it doesn't have to be in ComfyUI folders
🖼️ Image Preview Instant preview, no need to start the workflow. Can be toggled on/off
📊 A1111/Civitai metadata extraction
🔄 ComfyUI workflow extraction
Returns file path of loaded image

Parameters

Required Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | img_path | STRING | Direct path to an image file | - |

Outputs

| Parameter | Type | Description | |-----------|------|-------------| | IMAGE | IMAGE | The loaded image (from image input or widgets) | | FILE_PATH | STRING | Path of loaded file | | metadata | METADATA | A1111/Civitai metadata extracted from file | | workflow | WORKFLOW | ComfyUI workflow extracted from file |

</details>

Smart Image Loader 🖼

<details> <summary>Click to expand full documentation</summary>

A flexible image loader that supports multiple input sources with metadata extraction.

Features

Multiple sources: txt, direct path, directory, or direct image
Smart priority order with configurable inputs
Random selection with reproducible seed
Multiple formats supported (PNG, JPG, JPEG, BMP, WEBP, TIFF)
🖼️ Image Preview
📊 A1111/Civitai metadata extraction
🔄 ComfyUI workflow extraction
Returns file path of loaded image

Parameters

Required Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | seed | INT | Seed for reproducible random selection | 0 | | img_dir_level | INT | Subdirectory depth: -1=all, 0=current, 1-10=levels | 0 | | show_preview | BOOLEAN | Show image preview in the node | True |

Optional Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | txt_path | STRING | Path to text file containing image paths | - | | img_path | STRING | Direct path to an image file | - | | img_directory | STRING | Path to directory containing images | - | | image | IMAGE | Direct image input (priority for IMAGE output) | - |

Outputs

Priority Order

The node loads images in this order (from highest to lowest priority):

image ⚡ : Uses the provided image input (for passing images only - does not load metadata or workflow)
txt_path 📄 : Randomly selects a path from text file
img_path 🖼 : Loads the specific image file
img_directory 📁 : Randomly selects an image from directory

Usage Examples

Case 1: Text file with image list

# Content of image_list.txt:
# /path/to/image1.png
# /path/to/image2.jpg
# /path/to/image3.webp

txt_path = "/path/to/image_list.txt"
seed = 42  # Reproducible

Case 2: Direct path to image

img_path = "/path/to/specific/image.png"

Case 3: Image directory

img_directory = "/path/to/images/"
seed = 123

Case 4: Combination with priority

txt_path = "/path/to/list.txt"       # Priority 1
img_path = "/path/to/fallback.png"   # Priority 2 (if txt_path fails)
img_directory = "/path/to/backup/"   # Priority 3 (if img_path fails)

Technical Details

Text file format (txt_path)

The text file should contain one image path per line:

/home/user/images/photo1.png
/home/user/images/photo2.jpg
/home/user/images/photo3.webp

Empty lines are ignored
UTF-8 encoding supported
A random path is selected using the seed

Seed handling

seed = 0: Different random selection each run
seed > 0: Identical selection with same parameters (reproducible)

FILE_PATH values

| Value | Description | |-------|-------------| | Full path | Image loaded from txt_path, img_path or img_directory | | "external_input" | Image provided via image parameter | | "none" | No valid source found (error) |

</details>

Smart Image Saver 💾

<details> <summary>Click to expand full documentation</summary>

An intelligent image saver that replaces the SAVE_IMG subgraph with flexible filename customization.

Features

Flexible naming: Customizable prefix, suffix, and separator
Date tokens: YYYY, MM, DD, HH, mm, ss, timestamp support
Smart metadata extraction: Auto-extract seed and model name from workflow
Node targeting: Target nodes by class name or direct ID (#10)
Multiple formats: WEBP (lossy/lossless), PNG, JPG/JPEG
Metadata preservation: Saves prompt and workflow in EXIF/PNG info

Parameters

Required Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | images | IMAGE | Images to save | - | | output_folder | STRING | Output folder (supports date tokens) | YYYY-MM-DD | | prefix | STRING | Filename prefix (supports date tokens) | ComfyUI_YYYY-MM-DD_HHmmss | | seed_node_name | STRING | Node containing seed (class name or #ID) | KSampler | | model_node_name | STRING | Node containing model (class name or #ID) | CheckpointLoaderSimple | | suffix | STRING | Filename suffix (supports date tokens) | "" | | output_format | COMBO | Image format | webp | | quality | INT | Compression quality (1-100) | 97 |

Outputs

| Parameter | Type | Description | |-----------|------|-------------| | images | IMAGE | Pass-through of input images | | filepath | STRING | Path of the first saved file |

Filename Structure

The node builds filenames in this order:

[prefix]_[seed]_[model_name]_[suffix].[format]

Elements are joined by the separator. Empty elements are skipped.

Usage Examples

Case 1: Basic usage with date folder

output_folder = "YYYY-MM-DD"           # → 2024-01-30/
prefix = "render_YYYYMMDD"              # → render_20240130
seed_node_name = "KSampler"
model_node_name = "CheckpointLoaderSimple"

Output: 2024-01-30/render_20240130_123456_mymodel.webp

Case 2: Target node by ID

seed_node_name = "#10"                  # Target node with ID 10
model_node_name = "#5"                  # Target node with ID 5

Case 3: Disable seed or model

seed_node_name = ""                     # Don't include seed
model_node_name = ""                    # Don't include model
prefix = "my_render"
suffix = "HHmmss"                       # Add timestamp

Output: my_render_143025.webp

Date Tokens

All date tokens are replaced with current date/time values:

| Token | Description | Example | |-------|-------------|---------| | YYYY | Year (4 digits) | 2024 | | MM | Month (2 digits) | 01 | | DD | Day (2 digits) | 30 | | HH | Hour 24h (2 digits) | 14 | | mm | Minutes (2 digits) | 30 | | ss | Seconds (2 digits) | 25 | | timestamp | Unix timestamp | 1706626225 |

Node Targeting

By class name:

seed_node_name = "KSampler"
model_node_name = "CheckpointLoaderSimple"

By ID:

seed_node_name = "#10"                   # Targets node with ID 10

</details>

Smart Image Saver (Advanced) 💾

<details> <summary>Click to expand full documentation</summary>

Node for saving images with advanced file naming and A1111/Civitai Auto V3 compatible metadata.

Category: TOO-Pack/image

Features

Dynamic file naming with workflow extraction
Metadata injection/editing
Support for 3 universal inputs (any1, any2, any3)
Automatic model and LoRA extraction with hash calculation
Targeted text replacement
Customizable date formatting
A1111/Civitai compatible metadata
Output formats: PNG, JPG, WebP

Inputs

| Parameter | Type | Description | |-----------|------|-------------| | images | IMAGE | Images to save (required) | | metadata | METADATA | Optional metadata | | workflow | WORKFLOW | Workflow to embed | | any1 | * | Universal input 1 | | any2 | * | Universal input 2 | | any3 | * | Universal input 3 |

Outputs

| Output | Type | Description | |--------|------|-------------| | images | IMAGE | Images passthrough | | filepath | STRING | Saved file path |

Configuration (JS Interface)

DATA

Create custom data fields for metadata and naming.

Value formats:

Static text: "my_text"
Widget extraction: #123:widget_name (node ID) or ClassName:widget_name
Any input: [any1], [any2], [any3]

Example:

name: positive
value: #45:text

name: model_used
value: [any1]

DATE FORMAT

Format dates for naming.

date1: Date format (e.g. YYYY-MM-DD)
date2: Time format (e.g. HHmmss)
date3: Custom format

Available tokens: YYYY, YY, MM, DD, HH, mm, ss, timestamp

MODEL

Automatic model extraction.

Value: #123:ckpt_name or [any1]

The node automatically calculates:

Model name (basename without extension)
Model hash (first 10 characters of SHA256)

LORAS

Automatic LoRA extraction with Civitai-compatible hash.

Add multiple loras:

Lora 1: #45:lora_name or [any1]
Lora 2: #67:lora_name or [any2]

Multiline support: If [any1] contains:

lora1.safetensors
lora2.safetensors

→ Both are parsed automatically

TEXT REPLACE

Replace text in fields before naming.

Target:

Empty or [any1]/[any2]/[any3] → Apply to all fields
positive → Apply only to "positive" field
model → Apply only to model

Example:

target: positive
in: (masterpiece)
out:

→ Removes "(masterpiece)" from positive prompt

NAMING

Build filename with available elements.

Available fields:

output_folder : Output subfolder
prefix : File prefix
extra1, extra2, extra3 : Additional fields
model : Model name
suffix : File suffix
separator : Separator character (default: _)

Available sources:

Empty (ignored)
[any1], [any2], [any3] : Universal inputs
Data field name: positive, seed, steps, etc.
model, loras : Extracted values
%date1, %date2, %date3 : Formatted dates

Naming example:

prefix: %date1
extra1: positive
model: model
suffix: seed
separator: _

→ 2025-02-12_beautiful_landscape_MyModel_12345.webp

OUTPUT

Output configuration.

format: png, jpg, webp
quality: 1-100 (for jpg/webp)
save metadata: Include A1111 metadata
embed workflow: Embed ComfyUI workflow (except JPG)

Usage Examples

Example 1: Naming with model and seed

Configuration:

DATA:
  - name: seed, value: #10:seed

MODEL:
  - extract: KSampler:ckpt_name

NAMING:
  - prefix: %date1
  - model: model
  - suffix: seed
  - separator: _

Result: 2025-02-12_MyCheckpoint_8675309.webp

Example 2: LoRA extraction from any1

Workflow:

ExtractWidgetFromNode → any1 (TOO Smart Image Saver)

Configuration:

LORAS:
  - lora 1: [any1]

NAMING:
  - prefix: %date1
  - extra1: loras

If any1 contains:

style_lora_v2.safetensors
quality_lora_v1.safetensors

Generated metadata:

Lora hashes: "style_lora_v2: a1b2c3d4e5f6, quality_lora_v1: f6e5d4c3b2a1"

Example 3: Prompt cleanup with Text Replace

Configuration:

TEXT REPLACE:
  - target: positive
  - in: (masterpiece, best quality)
  - out: 

NAMING:
  - prefix: positive

If positive = "beautiful landscape, (masterpiece, best quality)"
→ Filename: beautiful_landscape.webp

Practical Use Cases

1. Multi-LoRA Workflow

Use ExtractWidgetFromNode to retrieve all used LoRAs → connect to any1 → The node automatically calculates all hashes for Civitai compatibility.

2. Organization by Model

Configure output_folder: model to automatically sort images by model used.

3. Complete Metadata

Create data fields for all important parameters (seed, steps, cfg, sampler, scheduler) → The node generates complete A1111 metadata readable by Civitai and Automatic1111.

4. Smart Naming

Use any inputs to inject dynamic information from other nodes (tags, descriptions, scores, etc.).

Tips

Data fields are flexible: create as many fields as needed
[any1], [any2], [any3] accept any data type (converted to string)
Multiline is supported for loras: one line = one lora
LoRA hashes exclude safetensors metadata (Civitai AutoV3 compatible)
Text replace applies to extracted values before filename construction
Empty lines in data fields are automatically ignored
The metadata input has priority over node data fields, unless data fields are explicitly set (enables metadata injection/editing)

</details>

Extract Widget From Node 🔧

<details> <summary>Click to expand full documentation</summary>

Extracts specific widget values from any node in the ComfyUI workflow.

Features

Targeted extraction of specific widgets by name
Multiple widgets supported (comma-separated)
Multi-node support: Targets all nodes matching class name
Direct ID targeting with #ID syntax
Intelligent parsing of nested structures
"on" filter: Ignores disabled elements

Parameters

Required Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | node_name | STRING | Node class name or #ID to target | - | | widget_names | STRING | Widget name(s) to extract (comma-separated) | - |

Outputs

| Parameter | Type | Description | |-----------|------|-------------| | extracted_values | STRING | Extracted values (multiline if multiple) |

Usage Examples

Case 1: Extract a single widget

node_name = "KSampler"
widget_names = "seed"

Output: "12345"

Case 2: Extract multiple widgets

node_name = "CLIPTextEncode"
widget_names = "text, strength"

Output:

beautiful landscape, masterpiece
0.85

Case 3: Target specific node by ID

node_name = "#10"
widget_names = "ckpt_name"

Case 4: Extract from multiple matching nodes

node_name = "LoraLoader"
widget_names = "lora_name"

Output (if 2 LoraLoader nodes exist):

style_lora.safetensors
detail_lora.safetensors

Technical Details

Node Targeting

The node supports two targeting modes:

By class name (e.g., "KSampler")
- Finds all nodes of this class
- Extracts from all matching nodes
By ID (e.g., "#10")
- Targets specific node with ID 10
- Only one node targeted

Widget Extraction

Simple values: "my_value"
Multiple values: Separated by newlines
Comma support: "widget1, widget2" extracts both

Nested Dictionary Handling

The node intelligently handles nested dictionaries:

Simple structure:

{
  "lora_name": "my_lora.safetensors",
  "strength_model": 0.85
}

Nested structure (Power Lora Loader):

{
  "loras": {
    "on": true,
    "lora_name": "my_lora.safetensors",
    "strength_model": 0.85
  }
}

The node automatically extracts values from both structures.

"on" filter

If a dictionary contains "on": false, its values are ignored.

Advanced Use Cases

1. Extract prompts from workflow

node_name = "CLIPTextEncode"
widget_names = "text"

2. Retrieve used seeds

node_name = "KSampler"
widget_names = "seed"

3. List all loaded models

node_name = "CheckpointLoader"
widget_names = "ckpt_name"

4. Extract control parameters

node_name = "ControlNetLoader"
widget_names = "control_net_name, strength"

</details>

TOO Crop Image 🖼

<details> <summary>Click to expand full documentation</summary>

Interactive tool for cropping your image before output.

If you use the image input then the cropping will apply to it. However the interactive preview works only with img_path input so you will need to use that. You can also use only img_path like you would with a Simple Image Loader

</details>

TOO LoRA Grid 🧪

<details> <summary>Click to expand full documentation</summary>

Modified KSampler to test loras and output an image grid.

A ComfyUI node to test multiple LoRA files and compare their effects in an image grid, with optional labels and footer text.

Inputs (high-level)

model: MODEL
clip: CLIP
vae: VAE
positive: CONDITIONING
negative: CONDITIONING
latent_image: LATENT — used as the sampling starting point
loras: STRING (multiline) — one entry per line (see format below)
seed, steps, cfg (float guidance), sampler_name, scheduler, denoise
grid_cols, grid_padding — grid layout controls
add_labels (bool), label_height, font_size, label_color, bg_color
footer_text — optional footer displayed beneath grid

LoRA multiline format

Each non-comment line is one grid cell. Lines starting with # are ignored.

Supported entry forms:

path/to/lora.safetensors
Apply LoRA at default strength 1.0 (if no explicit weight).
path/to/lora.safetensors:0.8
Apply LoRA at explicit strength (float).
path/to/lora.safetensors:"label"
A null slot — the node will sample WITHOUT applying any LoRA but will show the custom label. Useful for control comparisons.

Notes:

If extension is omitted, the node will try common extensions: .safetensors, .pt, .ckpt.
You may include a filename without path; the node resolves via project LoRA folders (uses folder_paths.get_full_path("loras", ...)).

Example loras widget content:

any_lora.safetensors:"no lora"
# test LoRAs
my_style_lora.safetensors:0.7
other_lora:0.5
missing_lora  # will be logged as not found

Grid & label rendering

Labels and footer use PIL ImageFont:
- The node tries these fonts (in order): arial.ttf, DejaVuSans.ttf, LiberationSans-Regular.ttf.
- If none load, it falls back to ImageFont.load_default() (Pillow's bitmap font).
Footer text is wrapped to grid width and supports multi-line content.
Colors accept hex strings (e.g., #ffffff for white, #111111 for dark background).
grid_cols controls number of columns; rows are computed automatically.

Caching

The node caches per-entry decoded tensors and labels in an in-memory _cache keyed by a hash of:
- model/clip/vae/positive/negative object signatures, loras text, seed, steps, cfg, sampler/scheduler, denoise, and a short hash of the latent image bytes.
Cache avoids re-sampling identical requests; change any parameter or the latent image and a new sample will be produced.

Tips

Use null slots (:"label") to include a no-LoRA baseline in the grid for direct comparison.
Use has_explicit_weight (via :0.8) to show exact strength in labels.
Keep grid_cols reasonable for readability; increase padding or label height if text overlaps images.
If you run ComfyUI in a container, ensure fonts are installed for consistent label rendering.

Limitations

One sample per LoRA entry — not intended for batch/randomized comparisons unless you feed different latents/seeds.
In-memory caching will not persist across process restarts.
The node tries to resolve LoRA paths via the project's LoRA folders; non-standard locations may not be found.

</details>

TOO Image Grid From Batch

<details> <summary>Click to expand full documentation</summary>

Creates an image grid from a batch while preserving transparency (alpha channel).

Parameters

Images (required)

The batch of images to arrange in a grid. Format: RGBA or RGB.

Grid Cols (required)

Number of columns in the grid.

Grid Rows (required)

Number of rows in the grid.

Padding (required)

Space in pixels between images. Min: 0 | Max: 100 | Default: 0

Output

A single image containing the grid composed of the batch images.

Notes

If the batch contains fewer images than the grid (rows × cols), missing cells are filled with transparency.
Transparency is fully preserved.
The batch images must have the same dimensions.

</details>

TOO Image Metadata 🖼

<details> <summary>Click to expand full documentation</summary>

Editing tool for A1111/Civitai metadata.

Very straight forward, you will need TOO Simple Image Loader or TOO Smart Image Loader to input the metadata to modify. Any empty field will be ignored, the node will not erase any original metadata unless you specify something. For the model field you might want to follow A1111/Civitai syntax [Model hash: <hash>, Model: modelName] -> Only 1 model can be input, as far as I know... Maybe something to investigate later x) As for the loras, the syntax is [Lora hashes: "<lora1>:<hash>,<lora2>:<hash>,<lora....>:<hash>"] -> As many loras as necessary

</details>

Krita Bridge 🎨

<details> <summary>Click to expand full documentation</summary>

Automatically loads the latest image from the input/krita/ folder for seamless Krita integration.

Features

Automatic loading of latest image from krita folder
Real-time detection of new files
Auto-update during generation
Alpha support: extracts alpha channel as mask
Auto-create folder if doesn't exist
No parameters - works directly

Parameters

Required Parameters

No parameters required - The node works automatically!

Outputs

| Parameter | Type | Description | |-----------|------|-------------| | image | IMAGE | The loaded image (RGB) | | mask | MASK | Mask extracted from alpha channel |

Setup

Working Directory

The node automatically loads from:

ComfyUI/input/krita/

If the folder doesn't exist, it's created automatically on first launch.

Export from Krita

Open Krita
Create or modify your image
Export as PNG: File > Export
Destination: ComfyUI/input/krita/filename.png
The node automatically detects the new file

Recommended Workflow

Krita → Export PNG → ComfyUI/input/krita/ → Krita Bridge → [your workflow]

Usage Examples

Case 1: Simple loading

# No configuration needed
# Node automatically loads latest image

Case 2: Using alpha mask

# Connect 'mask' to a mask node (Mask Composite, etc.)
# Krita image's alpha channel becomes usable mask

Case 3: Iterative Krita ↔ ComfyUI workflow

1. Draw in Krita
2. Export → input/krita/sketch.png
3. ComfyUI detects and generates
4. Get result
5. Refine in Krita
6. Re-export → Node loads new version

Case 4: Inpainting with Krita mask

# Workflow:
# Krita Bridge (image + mask) → Inpaint Model → VAE Decode
# Krita's alpha mask defines inpainting area

Technical Details

File Detection

The node:

Scans input/krita/ folder for all .png files
Finds file with most recent modification
Compares timestamp with last loaded file
Reloads if change detected

Alpha Channel Handling

RGBA image (with transparency):

RGB channels → image output
Alpha channel → mask output (0-1 values)

RGB image (no transparency):

RGB → image output
Uniform white mask → mask output

Automatic Update

The IS_CHANGED function returns current timestamp, forcing ComfyUI to:

Re-evaluate node on every execution
Detect new files in real-time
Update image automatically

Krita Integration

Krita Configuration

Set default export folder
- Settings > Configure Krita > General
- Set default folder: ComfyUI/input/krita/
Keyboard shortcut for quick export
- Settings > Configure Krita > Keyboard Shortcuts
- Assign key to Export
- Example: Ctrl+Shift+E
Export format
- Format: PNG
- Compression: as preferred
- Important: Enable "Save alpha channel" if using mask

Optimal Workflow

┌─────────┐    Export PNG    ┌──────────────┐
│  Krita  │ ───────────────> │ input/krita/ │
└─────────┘                   └──────────────┘
                                      │
                                      ▼
                              ┌──────────────┐
                              │ Krita Bridge │
                              └──────────────┘
                                      │
                     ┌────────────────┴────────────────┐
                     ▼                                 ▼
                ┌────────┐                        ┌──────┐
                │ image  │                        │ mask │
                └────────┘                        └──────┘

</details>

Collection Categorizer (LLM) 🗂

<details> <summary>Click to expand full documentation</summary>

A ComfyUI node that automatically scans your folders and categorizes content with a local LLM (Ollama).

Features

Automatic scanning of folders (video files, archives, documents)
Smart categorization via local LLM (Ollama)
Recursive scanning of subfolders (optional)
Reproducible seed for identical results
Custom Ollama models supported
Auto-save JSON output
Compatible with Collection Manager
100% local - no external APIs

Parameters

Main Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | ollama_model | COMBO | LLM model to use (or "custom") | qwen2.5:7b | | custom_ollama_model | STRING | Model name if "custom" selected | - | | folder_path | STRING | Path to folder to scan | - | | scan_subfolders | BOOLEAN | Recursively scan subfolders | False | | save_json | BOOLEAN | Auto-save JSON output | True | | collection_title | STRING | Collection title | Ma Collection | | content_type | COMBO | Content type (or "custom") | films |

Optional Parameters

| Parameter | Type | Description | Default | |-----------|------|-------------|---------| | custom_type_name | STRING | Type name if "custom" selected | - | | custom_categories | STRING | Custom categorization criteria (multiline) | (empty = LLM decides) | | seed | INT | Seed for reproducible results | 0 (random) |

Outputs

| Parameter | Type | Description | |-----------|------|-------------| | json_output | STRING | The complete collection JSON | | summary | STRING | Categorization summary |

Usage Examples

Case 1: Movies with automatic categories

content_type = "films"
custom_categories = ""  # Empty
# LLM decides categories (Genre, Era, etc.)

Case 2: TV Series with custom criteria

content_type = "series"
custom_categories = "Genre, Year, Studio"
# LLM categorizes by these criteria

Case 3: Custom type

content_type = "custom"
custom_type_name = "Documentaries"
custom_categories = "Theme, Duration"

Case 4: Custom model

ollama_model = "custom"
custom_ollama_model = "mistral:7b"

Case 5: Reproducible results

seed = 42
# Always same result with same parameters

Installation

Requirements

Ollama installed and running

# Download: https://ollama.ai
ollama --version

Python requests (for local HTTP API)

pip install requests --break-system-packages

At least one LLM model
```
ollama pull qwen2.5:7b
```

Supported File Types

Videos .mp4, .mkv, .avi, .mov, .wmv, .flv

Archives .cbz, .cbr, .zip, .rar

Documents .epub, .pdf, .mobi

Folders Subfolders are treated as individual items (unless scan_subfolders is enabled)

Recommended Models

| Model | Size | Speed | Quality | Usage | |-------|------|-------|---------|--------| | qwen2.5:7b | 7B | ⚡⚡⚡ | ⭐⭐⭐⭐ | Recommended | | gemma3:12b | 12B | ⚡⚡ | ⭐⭐⭐⭐⭐ | Best quality | | llama3.1:8b | 8B | ⚡⚡⚡ | ⭐⭐⭐⭐ | Very reliable | | gemma3:4b | 4B | ⚡⚡⚡⚡ | ⭐⭐⭐ | Fast |

Output Format (JSON)

{
  "title": "My Collection",
  "icon": "🎬",
  "type": "Films",
  "filename": "films.json",
  "categories": [
    {
      "id": 1,
      "name": "Science Fiction",
      "subcategories": [],
      "games": ["Blade Runner", "The Matrix"]
    },
    {
      "id": 2,
      "name": "Comedy",
      "subcategories": [],
      "games": ["Superbad", "The Hangover"]
    }
  ]
}

</details>

Time Converter 🕥

<details> <summary>Click to expand full documentation</summary>

A Quick tool for converting seconds/min/hours.

🗺 ToDo

Extract Widget From Node : aiming at nodes using name, title or ID

📄 License

MIT License

👤 Author

Tetsuoo

🙏 Credits

Claude AI ❤ - AI assistant extraordinaire
ComfyUI - Amazing node-based interface
Ollama - Local LLM runtime
Krita - Open-source digital painting software