Text Encoder Inspector (Qwen3-4B)
Qwen3_4BTextEncoderInspector
model_patches/analysisoutput node
Analyzes Qwen3-4B text encoder layers. ⚡ QUICK MODE (default): Fast, low VRAM - Captures activation magnitudes per layer - Shows which layers respond strongest to your prompt ⚠️ ABLATION MODE: Slow, HIGH VRAM (3x model size!) - Tests what happens when each region is weakened - Gives precise importance scores Use QUICK mode first, then ABLATION if needed.
Pack: Realtime LoRA Trainer
custom_nodes.comfyUI-Realtime-Lora