LoRA Tag Power Loader
Power LoRA loader that detects LoRA tags in text input. Supports dual noise weighting for WanVideo/Hunyuan models. Tag Format: - <lora:name:weight> - Sets model weight for both high/low noise. CLIP uses default_weight. - <lora:name:high:low> - Different model weights for high/low noise passes. CLIP uses default_weight. - <lora:name:high:low:clip> - Full control: separate high/low model weights and explicit CLIP weight. Features: - Text-based tag detection (like ImpactWildcardEncode) - Unlimited LoRA loading (like Power Lora Loader) - Dual noise weight support (high/low noise passes) - WanVideo/Hunyuan compatibility with video_model_mode - LoRA caching for performance - Global weight multiplier for quick adjustments Noise Mode: - high_noise: Use high noise weights (for early generation passes) - low_noise: Use low noise weights (for detail refinement passes) - auto: Average of high and low weights For WanVideo/Hunyuan dual LoRA loading, use two instances of this node with different noise_mode settings.
- model
- clip
- MODEL
- CLIP
- text
- lora_info
- trigger_words
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| text | STRING | Text containing LoRA tags like <lora:name:weight>. Tags are removed from output (or replaced with trigger words if auto_trigger is ON). Supports: <lora:name:weight>, <lora:name:high:low>, <lora:name:high:low:clip>. | |
| model | MODEL | The model to apply LoRAs to. | |
| noise_mode | COMBO | auto | 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. |
| clipopt | CLIP | CLIP to apply LoRAs to. Optional - if not provided, only model LoRAs will be applied. | |
| default_weightopt | FLOAT | 1.00-10–10 | Default weight for CLIP and for tags without weights. Model weights are set explicitly in tags. |
| weight_multiplieropt | FLOAT | 1.000–10 | Global multiplier for ALL LoRA weights. 2.0 doubles all weights, 0.5 halves them. Applied after individual tag weights. |
| video_model_modeopt | BOOLEAN | false | Enable LoRA key standardization for WanVideo/Hunyuan models. Turn ON when using video models. |
| auto_triggeropt | BOOLEAN | false | Extract trigger words from LoRA metadata and insert them where the tag was. Supports Civitai, Kohya, AI Toolkit, SimpleTuner, OneTrainer. Prevents LoRA tags from appearing as artifacts in images. |
Outputs (5)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | The model with LoRAs applied |
| CLIP | CLIP | The CLIP with LoRAs applied (if CLIP input was provided) |
| text | STRING | The input text with LoRA tags removed (or replaced with trigger words if auto_trigger is ON) |
| lora_info | STRING | Formatted list of loaded LoRAs with their weights, triggers, and any errors |
| trigger_words | STRING | Comma-separated list of all trigger words extracted from loaded LoRAs |