- Add extract_trigger_words() method supporting Civitai, Kohya, AI Toolkit, SimpleTuner, OneTrainer
- Add auto_trigger toggle to inject trigger words at LoRA tag positions
- Add trigger_words output socket with all extracted triggers
- Enhanced lora_info output to show trigger words per LoRA
- Fix duplicate LoRA tags causing rendering issues (v1.4.1)
- Update README with new examples and documentation
- Update CHANGELOG with v1.4.1 and v1.5.0 entries
A powerful ComfyUI custom node that combines text-based LoRA tag detection
with advanced power loading and dual noise support for WanVideo/Hunyuan models.
Features:
- Text-based LoRA tag parsing from prompts
- Dual noise weighting (high/low noise passes)
- Flexible tag format: <lora:name:weight>, <lora:name:high:low>, <lora:name:high:low:clip>
- Single weight affects model only; CLIP uses default_weight parameter
- LoRA caching for performance
- Global weight multiplier
- WanVideo/Hunyuan key standardization support
Tag behavior:
- <lora:name:1> sets model weight to 1.0, CLIP uses default_weight
- <lora:name:1:0.5> sets high=1.0, low=0.5 for model, CLIP uses default_weight
- <lora:name:1:0.5:0.7> sets high=1.0, low=0.5 for model, clip=0.7 explicitly
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
jonstreeter