ModelMergeCosmosPredict2_2B - ComfyUI Built-in Node Documentation

Complete documentation for the ModelMergeCosmosPredict2_2B node in ComfyUI. Learn its inputs, outputs, parameters and usage.

This documentation was AI-generated. If you find any errors or have suggestions for improvement, please feel free to contribute! Edit on GitHub

The ModelMergeCosmosPredict2_2B node merges two diffusion models using a block-based approach with fine-grained control over different model components. It allows you to blend specific parts of two models by adjusting interpolation weights for position embedders, time embedders, transformer blocks, and final layers. This provides precise control over how different architectural components from each model contribute to the final merged result.

Inputs

Parameter	Data Type	Required	Range	Description
model1	MODEL	Yes	-	The first model to merge
model2	MODEL	Yes	-	The second model to merge
pos_embedder.	FLOAT	Yes	0.0 - 1.0	Position embedder interpolation weight (default: 1.0)
x_embedder.	FLOAT	Yes	0.0 - 1.0	Input embedder interpolation weight (default: 1.0)
t_embedder.	FLOAT	Yes	0.0 - 1.0	Time embedder interpolation weight (default: 1.0)
t_embedding_norm.	FLOAT	Yes	0.0 - 1.0	Time embedding normalization interpolation weight (default: 1.0)
blocks.0.	FLOAT	Yes	0.0 - 1.0	Transformer block 0 interpolation weight (default: 1.0)
blocks.1.	FLOAT	Yes	0.0 - 1.0	Transformer block 1 interpolation weight (default: 1.0)
blocks.2.	FLOAT	Yes	0.0 - 1.0	Transformer block 2 interpolation weight (default: 1.0)
blocks.3.	FLOAT	Yes	0.0 - 1.0	Transformer block 3 interpolation weight (default: 1.0)
blocks.4.	FLOAT	Yes	0.0 - 1.0	Transformer block 4 interpolation weight (default: 1.0)
blocks.5.	FLOAT	Yes	0.0 - 1.0	Transformer block 5 interpolation weight (default: 1.0)
blocks.6.	FLOAT	Yes	0.0 - 1.0	Transformer block 6 interpolation weight (default: 1.0)
blocks.7.	FLOAT	Yes	0.0 - 1.0	Transformer block 7 interpolation weight (default: 1.0)
blocks.8.	FLOAT	Yes	0.0 - 1.0	Transformer block 8 interpolation weight (default: 1.0)
blocks.9.	FLOAT	Yes	0.0 - 1.0	Transformer block 9 interpolation weight (default: 1.0)
blocks.10.	FLOAT	Yes	0.0 - 1.0	Transformer block 10 interpolation weight (default: 1.0)
blocks.11.	FLOAT	Yes	0.0 - 1.0	Transformer block 11 interpolation weight (default: 1.0)
blocks.12.	FLOAT	Yes	0.0 - 1.0	Transformer block 12 interpolation weight (default: 1.0)
blocks.13.	FLOAT	Yes	0.0 - 1.0	Transformer block 13 interpolation weight (default: 1.0)
blocks.14.	FLOAT	Yes	0.0 - 1.0	Transformer block 14 interpolation weight (default: 1.0)
blocks.15.	FLOAT	Yes	0.0 - 1.0	Transformer block 15 interpolation weight (default: 1.0)
blocks.16.	FLOAT	Yes	0.0 - 1.0	Transformer block 16 interpolation weight (default: 1.0)
blocks.17.	FLOAT	Yes	0.0 - 1.0	Transformer block 17 interpolation weight (default: 1.0)
blocks.18.	FLOAT	Yes	0.0 - 1.0	Transformer block 18 interpolation weight (default: 1.0)
blocks.19.	FLOAT	Yes	0.0 - 1.0	Transformer block 19 interpolation weight (default: 1.0)
blocks.20.	FLOAT	Yes	0.0 - 1.0	Transformer block 20 interpolation weight (default: 1.0)
blocks.21.	FLOAT	Yes	0.0 - 1.0	Transformer block 21 interpolation weight (default: 1.0)
blocks.22.	FLOAT	Yes	0.0 - 1.0	Transformer block 22 interpolation weight (default: 1.0)
blocks.23.	FLOAT	Yes	0.0 - 1.0	Transformer block 23 interpolation weight (default: 1.0)
blocks.24.	FLOAT	Yes	0.0 - 1.0	Transformer block 24 interpolation weight (default: 1.0)
blocks.25.	FLOAT	Yes	0.0 - 1.0	Transformer block 25 interpolation weight (default: 1.0)
blocks.26.	FLOAT	Yes	0.0 - 1.0	Transformer block 26 interpolation weight (default: 1.0)
blocks.27.	FLOAT	Yes	0.0 - 1.0	Transformer block 27 interpolation weight (default: 1.0)
final_layer.	FLOAT	Yes	0.0 - 1.0	Final layer interpolation weight (default: 1.0)

Outputs

Output Name	Data Type	Description
model	MODEL	The merged model combining features from both input models