ComfyUI Node
Anima Multi-Model Merger
A ComfyUI node in HaDeS/Anima Mergers with 14 inputs and 2 outputs.
Anima Multi-Model Merger
- model_1
- model_2
- base_model
- model_3
- model_4
- MODEL
- info
◄methodTIES►
◄density0.50►
◄scale1.00►
◄seed0►
◄weight_11.00►
◄weight_21.00►
◄weight_31.00►
◄weight_41.00►
◄protect_llm_adaptertrue►
CategoryHaDeS/Anima Mergers
Inputs (14)
| Name | Type | Default | Description |
|---|---|---|---|
| model_1 | MODEL | — | |
| model_2 | MODEL | — | |
| method | COMBO | TIES | TIES: trim each task vector to its largest entries, elect a sign per entry, average the agreeing ones. DARE: randomly drop entries and rescale the rest (DARE-TIES then elects signs like TIES). Task Arithmetic: weighted average of the task vectors. Model Stock: average pulled toward base_model by the angle between the task vectors (ignores weights, density and scale). SLERP: spherical blend of model_1 and model_2 only, t = weight_2 / (weight_1 + weight_2); needs no base_model. |
| density | FLOAT | 0.500.05–1 | TIES: fraction of each task vector kept (largest entries). DARE: probability of keeping each entry. |
| scale | FLOAT | 1.000–2 | Strength of the merged task vector on top of base_model. 1.0 has the strength of an average; above 1.0 extrapolates. |
| seed | INT | 00–18446744073709550000 | DARE drop mask. The same seed gives the same merge on the same kind of device. |
| weight_1 | FLOAT | 1.000–2 | Relative weight of model_1 (normalized by the sum of the connected models' weights). |
| weight_2 | FLOAT | 1.000–2 | Relative weight of model_2 (normalized by the sum of the connected models' weights). |
| base_modelopt | MODEL | The checkpoint the models were finetuned from (e.g. anima-base-v1.0). Required by every method except SLERP. | |
| model_3opt | MODEL | — | |
| model_4opt | MODEL | — | |
| weight_3opt | FLOAT | 1.000–2 | Relative weight of model_3 (normalized by the sum of the connected models' weights). |
| weight_4opt | FLOAT | 1.000–2 | Relative weight of model_4 (normalized by the sum of the connected models' weights). |
| protect_llm_adapteropt | BOOLEAN | true | Keep the llm_adapter from base_model (from model_1 for SLERP), per Anima's llm_adapter_lr=0 rule. |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |
| info | STRING | — |