ComfyUI-hybs-nodes

Various custom nodes will be added.

Nodes

Resolution Selector

Provides a manual resolution selection node via a dropdown menu. Allows manual selection of a resolution from config/resolution_combos.json (or the default list).

• Category: HYBS/ResolutionSelector

• Inputs:

  • resolution (Combo dropdown): A string in the format "<width>x<height>".

• Outputs:

  • width (INT)
  • height (INT)

• Behavior:

  1. On startup, loads options from config/resolution_combos.json; if missing or invalid, falls back to default list.
  2. Populates the dropdown with "WxH" strings for each pair.
  3. Throws a runtime error if an unexpected format is selected.

Random Resolution Selector

Provides a dynamic resolution selection node based on a seed input.

Outputs a resolution (width, height) from a predefined list based on a seed input. It reloads config/resolution_combos.json on each execution, so changes to the JSON file take immediate effect.

• Category: HYBS/ResolutionSelector

• Inputs:

  • seed (INT): The seed value; same seed always yields the same resolution.

• Outputs:

  • width (INT)
  • height (INT)

• Behavior:

  1. On each node execution, attempts to load config/resolution_combos.json; if missing, uses the default hardcoded list.
  2. Throws a runtime error if the JSON file exists but is malformed or has invalid format.
  3. Determines the index with seed % len(combos) to select the resolution.

SeedListGenerator

Generates a list of random seed values.

• Category: HYBS/SeedGenerator

• Inputs:

  • count (INT): Number of random seeds to generate.

• Outputs:

  • seed list (LIST)
  • count (INT)

Conditional LoRA Loader

Applies one or more LoRAs only when the incoming positive prompt matches regex patterns defined in an external TOML file.

• Category: HYBS/ConditionalLoRALoader

• Inputs:

  • model (MODEL)
  • clip (CLIP)
  • positive (STRING): forceInput=True, multiline. Provide the positive prompt from another node.
  • config_toml (Combo dropdown): Select a .toml from config/.

• Outputs:

  • model (MODEL)
  • clip (CLIP)
  • applied loras (STRING)

• Behavior:

  1. On each execution, the selected TOML is read from config/. The node’s IS_CHANGED includes the file’s mtime, so saving the TOML triggers re‑execution on the next run.
  2. For each [[lora]] entry whose trigger (Python‑style regex) matches the positive prompt, the LoRA is applied.
  3. If multiple entries match, all are applied in order (top → bottom).
  4. LoRA loading tries, in order: comfy.sd.load_lora → built‑in nodes.LoraLoader → a low‑level fallback (if available). Logs show matched=..., Applied LoRA: ..., or errors.

• Important wiring:

  • Use the LoRA‑applied model/clip outputs from this node downstream.
  • Ensure CLIP Text Encode receives the LoRA‑applied clip (encode after applying LoRA). Otherwise the LoRA won’t affect conditioning.

TOML format

Place files in custom_nodes/ComfyUI-hybs-nodes/config/*.toml. Each file contains an array of [[lora]] entries:

# sample.toml — Conditional LoRA rules (English)
# Each [[lora]] entry is evaluated against the *positive* prompt using a Python-style regex.
# If multiple entries match, they are applied in order (top to bottom).
# `name` is a path relative to your `loras/` directory.

# Example 1: phrase “red dress” (case-insensitive), allowing one or more spaces
[[lora]]
trigger = "(?i)red\\s+dress"
name = "characters/wardrobe/red_dress_lora.safetensors"
strength_model = 1.0
strength_clip  = 1.0

# Example 2: either “nurse” or “white coat” (case-insensitive)
[[lora]]
trigger = "(?i)(nurse|white\\s*coat)"
name = "characters/nurse/nurse_v5.safetensors"
strength_model = 0.8
strength_clip  = 0.8

# Example 3: whole word “wizard” (case-insensitive)
[[lora]]
trigger = "(?i)\\bwizard\\b"
name = "styles/fantasy/wizard_style_v2.safetensors"
strength_model = 0.7
strength_clip  = 0.7

# Example 4: “short hair” but not “short hair cut” (case-insensitive, negative lookahead)
[[lora]]
trigger = "(?i)short\\s+hair(?!\\s*cut)"
name = "attributes/hair/short_hair_v1.safetensors"
strength_model = 0.6
strength_clip  = 0.6

Notes

• Regex is evaluated with Python’s re.search. If you want case‑insensitive matching, add the inline flag (?i) at the start of your pattern.
• Backslashes must be escaped in TOML strings (e.g., \s, \b → \\s, \\b).
• name must include the file extension and is resolved relative to your loras/ directory.

Configuration

Resolution lists (resolution_combos.json)

Place your custom resolution list in config/resolution_combos.json at the root of the extension (next to nodes/). The file must contain a non-empty JSON array of integer pairs, e.g.:

[
  [1024, 1024],
  [1088, 1088],
  [1152, 1152],
  [896, 1152],
  [832, 1216]
]

• If the file is missing, a built-in default list (1024–2048 squares and representative portrait/landscape) is used.
• If the file is present but invalid (empty array or incorrect format), the node will throw a value error.

Conditional LoRA rules (TOML)

Place TOML files under custom_nodes/ComfyUI-hybs-nodes/config/*.toml. Each file defines a list of rules using a [[lora]] array. For each entry:

• trigger: Python-style regex evaluated against the incoming positive prompt.
• name: LoRA file path relative to your loras/ directory (subfolders allowed; include the extension, e.g., .safetensors).
• strength_model, strength_clip: LoRA weights (default 1.0 if omitted).
• If multiple entries match, they are all applied in order (top → bottom).

Sample (config/sample.toml)

# Each [[lora]] entry is evaluated against the *positive* prompt using a Python-style regex.
# If multiple entries match, they are applied in order (top to bottom).
# `name` is a path relative to your `loras/` directory.

# Example 1: phrase “red dress” (case-insensitive), allowing one or more spaces
[[lora]]
trigger = "(?i)red\s+dress"
name = "characters/wardrobe/red_dress_lora.safetensors"
strength_model = 1.0
strength_clip  = 1.0

# Example 2: either “nurse” or “white coat” (case-insensitive)
[[lora]]
trigger = "(?i)(nurse|white\s*coat)"
name = "characters/nurse/nurse_v5.safetensors"
strength_model = 0.8
strength_clip  = 0.8

# Example 3: whole word “wizard” (case-insensitive)
[[lora]]
trigger = "(?i)\bwizard\b"
name = "styles/fantasy/wizard_style_v2.safetensors"
strength_model = 0.7
strength_clip  = 0.7

# Example 4: “short hair” but not “short hair cut” (case-insensitive, negative lookahead)
[[lora]]
trigger = "(?i)short\s+hair(?!\s*cut)"
name = "attributes/hair/short_hair_v1.safetensors"
strength_model = 0.6
strength_clip  = 0.6

Notes

• Regex is evaluated with Python’s re.search. To make it case-insensitive, add the inline flag (?i) to your pattern.
• Escape backslashes in TOML strings (e.g., \s, \b).
• Ensure your routing uses the LoRA-applied clip for Text Encode downstream, or the LoRA will not affect conditioning.

Installation

1. Clone this repository into your ComfyUI custom_nodes/ folder:

   cd path/to/ComfyUI/extensions
   git clone https://github.com/hybskgks28275/ComfyUI-hybs-nodes.git
   

2. Ensure the directory structure:

   ComfyUI/
   └── custom_nodes/
       └── ComfyUI-hybs-nodes/
           ├── config/
           │   ├── lora_condition.toml.example
           │   └── resolution_combos.json.example
           ├── nodes/
           │   ├── hybs_conditional_lora_loader.py
           │   ├── hybs_random_resolution_selector.py
           │   ├── hybs_resolution_selector.py
           │   └── hybs_seed_list_generator.py
           ├── hybs_resolution_common.py
           ├── LICENSE
           ├── README.md
           ├── __init__.py
           └── pyproject.toml
   

3. Restart ComfyUI. The nodes Resolution Selector, Random Resolution Selector will appear under HYBS/ResolutionSelector, and Conditional LoRA Loader will appear under HYBS/ConditionalLoRALoader.

License

This project is licensed under the MIT License. See LICENSE for details.