ComfyUI Node
LC Model Merge (any precision)
Merges two models stored in different precisions (bf16 with fp8, int8, nvfp4 or GGUF, in either order). The quantized one is decoded to full precision first and the full-precision model is the base, so the result stays full precision. Save it with LC Diffusion Model Save.
LC Model Merge (any precision)
- model1
- model2
- model
- report
◄ratio0.50►
◄block_ratios►
CategoryLC Modelbuilder
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| model1 | MODEL | Any precision: bf16, fp16, fp32, fp8, int8, nvfp4 or GGUF. | |
| model2 | MODEL | Same architecture as model1, any precision. | |
| ratio | FLOAT | 0.500–1 | 0 = all model1, 1 = all model2, 0.5 = half and half. |
| block_ratiosopt | STRING | Optional per-block ratios for models with numbered blocks (Krea 2 has 28): comma list, block 0 first, same scale as ratio. Layers outside the blocks use ratio. Empty = ratio everywhere. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| report | STRING | — |