ComfyUI Node
Corza Clean Applied LoRAs
Cleans standard LoRA patches already applied to the incoming MODEL. Place it after upstream LoRA loaders; downstream LoRA loaders remain untouched.
Corza Clean Applied LoRAs
- model
- model
◄keep_energy100.0►
◄max_rank0►
◄tame_layers0.00►
◄star_rescalefalse►
◄gate_strength1.00►
Categorycorza/lora
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| keep_energy | FLOAT | 100.050–100 | Per applied LoRA patch, keep only the strongest SVD components adding up to this % of the update's energy. 100 = off. |
| max_rank | INT | 00–1024 | Hard cap on each applied LoRA patch's rank after the energy cut. 0 = off. |
| tame_layers | FLOAT | 0.000–1 | Compress applied LoRA patches whose update is much stronger than the rest of the upstream LoRA stack. 0 = off, 1 = fully clamped. |
| star_rescale | BOOLEAN | false | STAR rescale after truncation. Only does something when keep_energy trims a patch or max_rank caps it. |
| gate_strength | FLOAT | 1.000–1 | How much of the applied LoRA patches' effect reaches 'gate' layers only (Krea 2 gated-attention + SwiGLU gates, etc.). Gates are multiplicative sigmoid controls whose LoRA edits have outsized, compounding effect — a big source of artifacts and of stacked LoRAs fighting. 1 = full effect (default), 0 = strip the LoRA from gates (they stay at base). Try 0.5 if stacked LoRAs deform. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |