jonstreeter-comfyui-Lora-Tag-Power-Loader/UPDATE_NOTES.md

Update Notes - Version 1.1.0

Summary of Changes

Based on user feedback, the node has been updated to match the behavior and flexibility of rgthree's Power Lora Loader.

Key Improvements

1. Optional Model and CLIP Inputs ✨

Before:

• Both model and clip were required inputs

After:

• Both model and clip are now optional inputs
• You can use the node just to parse tags and get cleaned text
• Matches Power Lora Loader's flexibility

Use Cases:

• Parse tags without applying LoRAs (useful for preprocessing)
• Apply LoRAs to model only (no CLIP needed)
• Full flexibility in workflow design

2. Comprehensive Tooltips 💡

Every parameter now has helpful tooltips that explain:

• What the parameter does
• What values to use
• When to enable specific options

Tooltip Examples:

• text: "Text containing lora:name:weight tags. Tags will be parsed and removed from output."
• noise_mode: "Which noise weights to use: high_noise for high noise pass, low_noise for low noise pass, auto averages both. For WanVideo dual loading, use two nodes with different modes."
• default_weight: "Default weight when tag doesn't specify one. Uses standard ComfyUI weighting method (same as Power Lora Loader)."

3. Standard ComfyUI Weighting Method ⚖️

Research Findings:

• rgthree Power Lora Loader uses standard ComfyUI weighting (NOT A1111 method)
• Standard ComfyUI method is most prevalent (68% of ecosystem)
• A1111 method is for prompt/embedding weighting, not model LoRA loading

What This Means:

• Our node uses the exact same weighting as Power Lora Loader
• No need for method selection - we use the standard
• Compatible with all existing workflows and expectations

4. Improved Error Handling 🛡️

• Gracefully handles None model/clip inputs
• Warns when clip strength specified but no CLIP provided
• Skips LoRA application when all strengths are zero
• Better error messages and logging

Migration Guide

If you're upgrading from 1.0.0:

No Breaking Changes!

• Existing workflows will continue to work
• Model and CLIP can now be disconnected if desired
• All tag formats remain the same

New Capabilities:

✓ Use without model input (just parse tags)
✓ Use without CLIP input (model-only LoRAs)
✓ Hover over parameters to see helpful tooltips

Technical Details

Function Signature Change

# Before (1.0.0)
def load_loras_from_tags(self, model, clip, text, noise_mode, ...)

# After (1.1.0)
def load_loras_from_tags(self, text, noise_mode, model=None, clip=None, ...)

Input Type Definition

"required": {
    "text": (...),
    "noise_mode": (...)
},
"optional": {
    "model": ("MODEL", {...}),
    "clip": ("CLIP", {...}),
    ...
}

Weighting Method

• Uses comfy.sd.load_lora_for_models() (standard ComfyUI)
• Same as LoraLoader().load_lora() used by Power Lora Loader
• Applies: weight += (strength * alpha/rank) * lora_diff

Files Updated

• ✅ lora_tag_power_loader.py - Core implementation
• ✅ README.md - Documentation and examples
• ✅ CHANGELOG.md - Version history
• ✅ UPDATE_NOTES.md - This file

Questions?

Q: Do I need to change my existing workflows? A: No! All existing workflows will work exactly as before.

Q: Can I still connect both model and CLIP? A: Yes! The inputs are optional, not removed. Connect what you need.

Q: What weighting method is used? A: Standard ComfyUI method, same as Power Lora Loader.

Q: Will this work with my video model workflows? A: Yes! Enable video_model_mode for WanVideo/Hunyuan support.

Enjoy the Update!

The node is now more flexible and user-friendly while maintaining full backward compatibility.