ComfyUI Node
LC Anima Block Merge Advanced
Anima merge with a draggable per-block ratio curve across the 28 blocks, plus sliders for the named parts outside them (x embedder, timestep embedder and norm, LLM adapter, final layer). Reuses ComfyUI's own ModelMergeBlocks. Save writes web/anima_block_presets/<name>.json.
LC Anima Block Merge Advanced
- model1
- model2
- model
◄preset50_50►
◄edit_modesmooth►
◄smooth_radius2►
◄blocks_curve0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5►
◄x_embedder0.50►
◄t_embedder0.50►
◄t_embedding_norm0.50►
◄llm_adapter0.50►
◄final_layer0.50►
◄save_name►
◄save_curvefalse►
CategoryLC ModelBuilder/Anima
Inputs (13)
| Name | Type | Default | Description |
|---|---|---|---|
| model1 | MODEL | Ratio 0 side. Blocks/regions at 0.0 come entirely from this checkpoint. | |
| model2 | MODEL | Ratio 1 side. Blocks/regions at 1.0 come entirely from this checkpoint. | |
| preset | COMBO | 50_50 | Bundled: 50_50, model2_early (blocks 0-6 from model2), model2_look (blocks 7-20, where an Anima finetune's look sits), model2_late (21-27), plus anything you save. Loading one sets the curve and the named sliders. Custom keeps what you dragged. |
| edit_mode | COMBO | smooth | smooth = drag one block and blend neighbors. spike = move one block only. |
| smooth_radius | INT | 20–8 | Neighbor window (in blocks) for smooth mode (0 = spike-like). |
| blocks_curve | STRING | 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5 | Comma list of 28 ratios (blocks.0. - blocks.27.). 0 = all model1, 1 = all model2. Source of truth when preset is Custom. |
| x_embedder | FLOAT | 0.500–1 | x_embedder. — turns the image latent into tokens, before block 0. |
| t_embedder | FLOAT | 0.500–1 | t_embedder. — timestep embedding. |
| t_embedding_norm | FLOAT | 0.500–1 | t_embedding_norm. — norm on the timestep embedding. |
| llm_adapter | FLOAT | 0.500–1 | llm_adapter. — maps the Qwen text encoder's output into the model (how the prompt is read). |
| final_layer | FLOAT | 0.500–1 | final_layer. — output layer, after block 27. |
| save_name | STRING | Filename under web/anima_block_presets/ (no extension). | |
| save_curve | BOOLEAN | false | On run, write save_name.json into web/anima_block_presets/. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |