ComfyUI-JK-TextTools

ComfyUI-JK-TextTools

Text and data manipulation nodes for ComfyUI, with emphasis on JSON processing, detection workflows, and bbox visualization.

Features

• Text Manipulation: Split, join, and index delimited strings with type casting
• JSON Processing: Format and query JSON data with wildcard filtering
• Detection Workflows: Extract and visualize bounding boxes from detection results
• Mask Generation: Convert bboxes and segmentations to masks for image processing
• Segmentation Support: Process SEGS format from SAM3 with filtering and union capabilities
• Mask to BBox Conversion: Convert masks back to bounding boxes for chaining workflows

Installation

Via ComfyUI Manager (Recommended)

(When published)

1. Open ComfyUI Manager
2. Search for "JK-TextTools"
3. Click Install

Manual Installation

1. Navigate to your ComfyUI custom_nodes directory:

   cd ComfyUI/custom_nodes
   

2. Clone this repository:

   git clone https://github.com/Nakamura2828/ComfyUI-JK-TextTools.git
   

3. Restart ComfyUI

Nodes

Text Manipulation

String Index Selector

Extract a single element from a delimited string by index - perfect for loop workflows.

Inputs:

• text (STRING): The delimited string to split
• delimiter (STRING): Character(s) to split on (default: ,)
• index (INT): Which item to extract (0-based by default)
• output_type (STRING/INT/FLOAT): Type to cast result to
• strip_whitespace (BOOLEAN): Remove leading/trailing spaces
• zero_indexed (BOOLEAN, optional): Use 0-based indexing

Outputs:

• selected_item: The extracted item (typed)
• item_count (INT): Total number of items

Example:

Input: "10,25,42,100", delimiter: ",", index: 2, output_type: INT
Output: 42 (as integer)

String Splitter

Split a delimited string into a typed list with optional casting.

Inputs:

• text (STRING): The delimited string
• delimiter (STRING): What to split on
• output_type (STRING/INT/FLOAT): Type to cast items to
• strip_whitespace (BOOLEAN): Clean up items
• remove_empty (BOOLEAN, optional): Remove empty strings

Outputs:

• string_list (LIST): List of typed items
• item_count (INT): Number of items

Features:

• Type casting to STRING, INT, or FLOAT
• Escape sequence support (\n, \t, \r)
• Empty string handling
• Grid icon displays correctly (OUTPUT_IS_LIST working)

List Index Selector

Extract an item from a list by index with type preservation.

Inputs:

• list_input (*): Any list (connect from String Splitter)
• index (INT): Which item to extract
• zero_indexed (BOOLEAN): 0-based or 1-based indexing

Outputs:

• selected_item: The selected item (type-preserving)
• list_length (INT): Total list size

String Joiner

Join list items into a delimited string.

Inputs:

• list_input (*): Any list
• delimiter (STRING): String to insert between items (supports escape sequences)

Outputs:

• joined_string (STRING): The combined string
• item_count (INT): Number of items joined

Escape Sequences: Supports \n, \t, \r, \\

JSON Processing

JSON Pretty Printer

Format JSON strings with proper indentation for readability.

Inputs:

• json_string (STRING): Raw JSON to format
• indent (INT): Number of spaces for indentation (0-8)
• sort_keys (BOOLEAN, optional): Alphabetically sort object keys

Outputs:

• formatted_json (STRING): Pretty-printed JSON
• is_valid (BOOLEAN): Whether JSON is valid
• error_message (STRING): Error details if invalid

Example:

Input: [{"detect_result":[{"class":"DOG","score":0.9}]}]
Output: (formatted with indentation and newlines)

Detection Query

Query detection results with class filtering, score thresholds, and wildcards.

Inputs:

• json_string (STRING): JSON containing detection results
• class_filter (STRING): Class name with wildcards (default: *)
• min_score (FLOAT, optional): Minimum confidence score
• max_results (INT, optional): Maximum results to return
• categorization_field (STRING, optional): Field name to extract

Outputs:

• filtered_json (STRING): Filtered results as JSON
• match_count (INT): Number of matches
• detection_list (LIST): Individual detections for iteration
• bbox_list (LIST): List of bboxes for visualization
• categorization_value (*): Extracted field value
• is_valid (BOOLEAN): Whether JSON is valid
• error_message (STRING): Error details if invalid

Wildcard Examples:

• CLASS1_LABEL → Exact match
• CLASS1_* → All CLASS1 subclasses
• *_LABEL → All ending with _LABEL
• * → All detections

Use Case: Filter detections, extract bboxes for visualization

BBox and Mask Operations

Detection to BBox

Extract bounding box from a detection object.

Inputs:

• detection (STRING): JSON string of detection object
• bbox_key (box/bbox): Which key contains the bbox

Outputs:

• bbox (BBOX): Bounding box in format [[x, y, width, height]]
• x, y, width, height (INT): Individual components
• class_name (STRING): Detection class
• score (FLOAT): Confidence score

Format: Works with detection objects containing "box": [x, y, w, h] or "bbox": [x, y, w, h]

JSON to BBox

Convert JSON bbox arrays to BBOX format with coordinate system conversion.

Inputs:

• json_string (STRING): JSON array of bboxes (e.g., from SAM3 Segmentation)
• input_format (XYXY/XYWH): Format of bboxes in JSON
• output_format (XYXY/XYWH): Format to output

Outputs:

• bboxes (LIST of BBOX): Converted bboxes in format [[[x,y,w,h]], ...]
• bbox_count (INT): Number of bboxes

Coordinate Formats:

• XYXY: Two corners [x1, y1, x2, y2] - used by SAM3 and other models
• XYWH: Corner + dimensions [x, y, width, height] - standard format

Example Input (SAM3 format):

[[245.3, 167.8, 512.6, 389.2], [100.0, 200.0, 300.0, 400.0]]

Use Case: Convert bbox output from nodes like TBG SAM3 Segmentation to work with mask generation nodes

BBox to Mask

Convert a single bounding box to a binary mask. Simple 1:1 conversion.

Inputs:

• bbox (BBOX): Single bbox [[x, y, w, h]]
• width (INT): Image width
• height (INT): Image height
• invert (BOOLEAN, optional): Invert mask (bbox black, rest white)

Outputs:

• mask (MASK): Binary mask for the bbox

Features:

• Simple single bbox → single mask conversion
• When connected to OUTPUT_IS_LIST sources, ComfyUI automatically iterates
• Handles both wrapped [[x,y,w,h]] and unwrapped [x,y,w,h] formats
• Automatic coordinate clamping to image bounds

Use Case: Single bbox to mask conversion. For multiple bboxes with union/combined mask, use BBoxes to Mask instead.

BBoxes to Mask ⭐ RECOMMENDED

Convert a list of bounding boxes to binary masks with union functionality.

Inputs:

• bboxes (*): List of bboxes from Detection Query or JSON to BBox
• width (INT): Image width
• height (INT): Image height
• invert (BOOLEAN, optional): Invert mask (bbox black, rest white)

Outputs:

• combined_mask (MASK): Union of all bboxes in one mask
• individual_masks (LIST of MASK): One mask per bbox
• bbox_count (INT): Number of bboxes processed

Features:

• Properly creates union mask (combined_mask) from multiple bboxes
• Individual masks for per-bbox processing
• Automatic coordinate clamping to image bounds
• Handles both wrapped and unwrapped bbox formats

Format: Accepts [[[x,y,w,h]], [[x,y,w,h]], ...] from Detection Query's bbox_list output or JSON to BBox

Mask to BBox ⭐ NEW

Convert a binary mask to a bounding box.

Inputs:

• mask (MASK): Binary mask tensor

Outputs:

• bbox (BBOX): Bounding box in format [[x, y, width, height]]
• x, y, w, h (INT): Individual integer coordinates

Features:

• Converts mask to tight bounding box
• Uses 0.5 threshold for float masks
• Floors all coordinates to integers
• Handles batched masks (uses first mask)
• Handles irregular shapes (L-shapes, non-convex masks)
• Returns [0,0,0,0] for empty masks

Use Case: Convert mask outputs (from detector nodes that output masks instead of proper BBOX type) to bbox format. Can chain output to other bbox nodes like BBox to SAM3 Query.

Example:

Mask (256x256)
  ↓
Mask to BBox
  ↓
bbox: [[50, 60, 100, 80]]
x: 50, y: 60, w: 100, h: 80 (all INT)
  ↓
BBox to SAM3 Query (chainable)

SEGs to Mask

Convert SEGS (segmentation results) to binary masks with filtering and union capabilities.

Inputs:

• segs (SEGS): Segmentation results from TBG SAM3 Segmentation or similar nodes
• label_filter (STRING, optional): Wildcard pattern for label filtering (default: *)
• min_confidence (FLOAT, optional): Minimum confidence threshold (0.0-1.0, default: 0.0)
• min_area_percent (FLOAT, optional): Minimum mask area as percentage of image (0.0-100.0, default: 0.0)
• sort_order (default/x_then_y/y_then_x/confidence_high_to_low, optional): Order segments deterministically
• union_same_labels (BOOLEAN, optional): Combine segments with same label (default: True)
• invert (BOOLEAN, optional): Invert mask (mask areas black, rest white)

Outputs:

• combined_mask (MASK): Union of all filtered segments
• individual_masks (LIST of MASK): One mask per label (when union enabled) or per segment
• labels_info (LIST of STRING): Label and confidence info (e.g., "person_0: 0.95")
• seg_count (INT): Number of masks returned

SEGS Format:

((height, width), [SEG(...), SEG(...), ...])

Each SEG object contains:

• cropped_mask: numpy array with mask data
• crop_region: [x1, y1, x2, y2] placement coordinates
• label: string label (e.g., "person_0")
• confidence: float confidence score

Features:

• Label Filtering: Use wildcards to filter by label (person_*, *_LABEL, etc.)
• Confidence Filtering: Remove low-confidence segments
• Area Filtering: Remove tiny masks below percentage threshold (e.g., 5.0 for 5% of image)
• Union Same Labels: Combines all segments with same label into one mask (default)
  • Uses max confidence score for combined segments
  • Example: 3x "person_0" segments → 1 "person_0" mask
• Deterministic Sorting: Order segments by position (x_then_y or y_then_x)
• Robust Handling: Gracefully handles None cropped_masks and invalid data

Wildcard Examples:

• * → All segments (default)
• person_* → All person detections
• *_0 → All first instances of each class
• dog → Exact match only

Sort Order Options:

• default → Keep original order from SEGS
• x_then_y → Left-to-right, then top-to-bottom
• y_then_x → Top-to-bottom, then left-to-right
• confidence_high_to_low → Sort by confidence score (highest first) ⭐ NEW

Use Case: Convert SAM3 segmentation results to masks for image processing workflows

SEGs to SAM3 Query ⭐ NEW

Convert SEGS segmentation format to SAM3 Selector query formats.

Inputs:

• segs (SEGS): Segmentation results from TBG SAM3 Segmentation or similar nodes

Outputs:

• box_query (STRING): JSON array with bounding box for SAM3 Selector [{"x1": 10, "y1": 20, "x2": 100, "y2": 150}]
• point_query (STRING): JSON array with weighted centroid [{"x": 55, "y": 85}]

Features:

• Generates both box and point queries for SAM3 Selector node
• Weighted centroid calculation for accurate point queries
• Handles multiple segments with union masks
• Coordinates clamped to image bounds
• Works with both ImpactPack and TBG SAM3 SEGS formats
• Handles numpy array and tensor masks
• Returns empty arrays [] for invalid inputs

Use Case: Chain SEGS segmentation output to SAM3 Selector for iterative refinement or re-segmentation.

Example Workflow:

TBG SAM3 Segmentation
  ↓
segs output → ((512, 512), [SEG(...), SEG(...), ...])
  ↓
SEGs to SAM3 Query
  ↓
box_query: [{"x1": 100, "y1": 200, "x2": 300, "y2": 400}]
point_query: [{"x": 200, "y": 300}]
  ↓
TBG SAM3 Selector
  - box_coords: box_query
  - point_coords: point_query
  ↓
Refined segmentation

Example Workflow:

TBG SAM3 Segmentation
    ↓
segs output → ((512, 512), [SEG(...), SEG(...), ...])
    ↓
SEGs to Mask
  - label_filter: "person_*"
  - min_confidence: 0.7
  - min_area_percent: 5.0
  - union_same_labels: True
    ↓
combined_mask → Mask of all detected persons
individual_masks → One mask per unique person
labels_info → ["person_0: 0.95", "person_1: 0.87"]

Workflow Examples

Example 1: Frame Number Extraction (Original Use Case)

VHS Node → "10,25,42,100"
    ↓
String Index Selector
  - delimiter: ","
  - index: 2 (from loop)
  - output_type: INT
    ↓
Output: 42 (as integer)
    ↓
Save Image: "frame_42.png"

Example 2: Detection Visualization

Detection JSON
    ↓
JSON Pretty Printer (format for readability)
    ↓
Detection Query
  - class_filter: "DOG_*"
  - min_score: 0.7
    ↓
bbox_list output → [[[x,y,w,h]], [[x,y,w,h]], ...]
    ↓
BBoxes to Mask
  - width: 512
  - height: 512
    ↓
combined_mask → Apply to original image
individual_masks → Process each detection separately

Example 2b: SAM3 Segmentation to Mask

TBG SAM3 Segmentation Node
    ↓
boxes output (JSON) → "[[x1,y1,x2,y2], [x1,y1,x2,y2], ...]"
    ↓
JSON to BBox
  - input_format: XYXY
  - output_format: XYWH
    ↓
bboxes output → [[[x,y,w,h]], [[x,y,w,h]], ...]
    ↓
BBoxes to Mask
  - width: 1024
  - height: 1024
    ↓
combined_mask → Apply to original image

Example 3: Typed List Processing

"10,25,42,100"
    ↓
String Splitter
  - output_type: INT
    ↓
[10, 25, 42, 100] (actual integers)
    ↓
List Index Selector
  - index: 2
    ↓
Output: 42 (as int, not string)
    ↓
Can connect directly to nodes expecting INT

Example 4: Multi-line Text Processing

Multiline Text Input
    ↓
String Splitter
  - delimiter: \n
    ↓
List of lines
    ↓
String Joiner
  - delimiter: ", "
    ↓
Comma-separated output

Development

Running Tests

# Individual tests
python tests/test_string_splitter.py

# All tests at once
python tests/run_all_tests.py

Requirements

• Python 3.10+
• ComfyUI
• PyTorch (for mask generation)

See requirements.txt for development dependencies.

Technical Notes

BBox Format

Standard format used throughout: [[x, y, width, height]]

• Single bbox: [[100, 200, 50, 75]]
• Multiple bboxes: [[[x1,y1,w1,h1]], [[x2,y2,w2,h2]], ...]
• Coordinates are XYWH (top-left corner + dimensions)

Alternative Coordinate Systems:

• XYXY: Two corners [x1, y1, x2, y2] - used by some detection models (SAM3, etc.)
• XYWH: Corner + dimensions [x, y, width, height] - our standard format
• Use JSON to BBox node to convert between these formats

OUTPUT_IS_LIST

Nodes using OUTPUT_IS_LIST show grid icon in ComfyUI and output items for iteration:

• String Splitter: string_list
• List Index Selector: receives lists
• Detection Query: detection_list, bbox_list
• BBoxes to Mask: individual_masks

Type Preservation

List Index Selector preserves input types:

• String list → Returns strings
• Int list → Returns ints
• Float list → Returns floats

Compatibility

Cross-Package Support

• Works with: ImpactPack (with type converter if needed)
• Works with: KJNodes BBox Visualizer
• Works with: TBG SAM3 Segmentation (via JSON to BBox node for bboxes, SEGs to Mask for segmentations)
• Works with: Standard ComfyUI mask nodes

Verified Workflows:

• TBG SAM3 → JSON to BBox → BBoxes to Mask ✅
• TBG SAM3 → SEGs to Mask (full SEGS support) ✅

Roadmap

Current Features ✅

• [x] Text splitting and joining with type casting
• [x] List indexing with type preservation
• [x] JSON formatting and validation
• [x] Detection querying with wildcards and bbox extraction
• [x] BBox extraction from detection objects
• [x] JSON bbox array conversion with XYXY/XYWH support
• [x] Mask generation from bboxes (union and individual)
• [x] SEGS segmentation to mask conversion with filtering and union
• [x] Escape sequence support
• [x] Comprehensive test suite (95%+ coverage)

Future Enhancements

• [ ] CSV Parser node
• [ ] JSONPath query support
• [ ] Regular expression nodes
• [ ] Additional bbox formats (XYXY, center-based)
• [ ] Mask operations (union, intersection, difference)
• [ ] String templating/formatting

License

MIT License

Author

John Knox (Nakamura2828)

Contributing

Issues and pull requests welcome!

Support

If you find these nodes useful, please star the repository on GitHub!

Changelog

v1.2.0 (2026-01-18)

• NEW: Mask to BBox Node - Convert masks to bounding boxes
  • Converts binary masks to BBOX format in XYWH coordinates
  • Outputs both BBOX type and individual x,y,w,h as INT values
  • Uses 0.5 threshold for float masks
  • Handles batched masks (uses first), irregular shapes, empty masks
  • All coordinates floored to integers
  • 12 comprehensive test cases
  • Use case: Convert mask outputs to bbox format for chaining to other nodes
• NEW: SEGs to SAM3 Query Node - Convert SEGS to SAM3 Selector queries
  • Generates box_query (bounding box) and point_query (centroid) for SAM3 Selector
  • Weighted centroid calculation for accurate point queries
  • Handles multiple segments with union masks
  • Full ImpactPack and TBG SAM3 compatibility
  • 12 comprehensive test cases covering all features
  • Use case: Chain SEGS segmentation to SAM3 Selector for iterative refinement
• ENHANCED: SEGs to Mask Node - Added confidence sorting
  • New sort_order option: "confidence_high_to_low"
  • Sorts output masks by confidence score (highest first)
  • Handles numpy array confidence values (ImpactPack compatibility)
  • Updated tests: 18 test cases (was 17)

v1.1.0 (2026-01-17)

• NEW: SEGs to Mask Node - Convert SEGS segmentation format to masks
  • Full support for TBG SAM3 Segmentation node output
  • Wildcard label filtering (fnmatch patterns)
  • Confidence threshold filtering (min_confidence)
  • Area percentage filtering (min_area_percent) to remove tiny masks
  • Deterministic sorting options (default/x_then_y/y_then_x)
  • Union same labels feature (combines segments with same label, uses max confidence)
  • Invert mode support
  • 4 outputs: combined_mask, individual_masks (list), labels_info (list), seg_count
  • Comprehensive tests: 16 test cases covering all features
  • Robust handling of None masks and invalid data

v1.0.1 (2026-01-17)

• BBox to Mask Refactored: Simplified to single bbox → single mask converter
  • Removed combined_mask output (use BBoxes to Mask for union functionality)
  • Renamed individual_masks output to just "mask"
  • Removed OUTPUT_IS_LIST - works as standard 1:1 converter
  • ComfyUI now iterates automatically when connected to list sources
  • Updated tests: 14 test cases validating simplified behavior
  • No longer marked EXPERIMENTAL - clean, focused implementation

v1.0.0 (2026-01-17)

• Initial release with 10 working nodes (now 11 with SEGs to Mask in v1.1.0)
• Text Manipulation (4 nodes):
  • String Index Selector, String Splitter, List Index Selector, String Joiner
  • Full type casting support (STRING/INT/FLOAT)
  • Escape sequence support
• JSON Processing (2 nodes):
  • JSON Pretty Printer with validation
  • Detection Query with wildcard filtering and bbox extraction
• BBox & Mask Operations (4 nodes):
  • Detection to BBox - Extract from detection objects
  • JSON to BBox - Convert JSON arrays with XYXY/XYWH conversion
  • BBoxes to Mask - Create union and individual masks (RECOMMENDED)
  • BBox to Mask - Simple 1:1 bbox to mask converter
• Comprehensive test suite (95%+ coverage)
• Full documentation (README.md + CLAUDE.md)
• Verified integrations with SAM3, KJNodes, ImpactPack