ComfyUI Node
TS LoRA Unmerged
Applies a LoRA as a side branch, y = Wx + BAx, instead of merging it into the weights. Merging loses a large part of a small LoRA on bf16 weights and adds noise on int8 — this keeps it whole, at the cost of roughly 10-25% per step. Made for few-step turbo LoRAs such as Viggle's for Qwen Image 2.1.
TS LoRA Unmerged
- model
- model
◄lora_name▾►
◄strength1.00►
CategoryTS/Utils
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | The diffusion model. It is not modified; a clone gets the LoRA. | |
| lora_name | COMBO | A LoRA of linear layers (diffusers/PEFT or kohya naming). | |
| strength | FLOAT | 1.00-100–100 | Multiplier on top of the file's own alpha / rank. Turbo LoRAs want 1.0. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | The same model with the LoRA added at run time. |