ModelMergeSD2
Same node as SD1, because SD2 has the same skeleton
- model1
- model2
- MODEL
Here's a fact that saves you an afternoon of confusion: in ComfyUI's source, ModelMergeSD2 is literally mapped to the same class as ModelMergeSD1, with a comment that reads "SD1 and SD2 have the same blocks." Same 30 sliders, same behavior, same node - the only difference is the label on the canvas. So if you came here from the SD2 node, read the ModelMergeSD1 article; everything there applies verbatim.
Which raises the honest question: why does this node exist at all? SD2.x is essentially dead. Stable Diffusion 2.0/2.1 shipped with a 768px native resolution, v-prediction, and a messy licensing situation, the fine-tune ecosystem that SD1.5 had never materialized around it, and the community's answer was to quietly go back to 1.5. A few years of model-generation history in miniature. You'd only be here if you loaded an SD2 checkpoint by accident, are archiving old files, or are working with the handful of legacy SD2 fine-tunes that still float around.
When you do use it, the mechanism is the family standard: clone model1 as the base, mix in model2's weights per block with ratio = how much of model2 you want. Same trap as every merge node - every slider defaults to 1.0, meaning "take model2," so a no-touch run gives you pure model2 rather than a blend. Same requirement that both inputs be the same architecture (SD2 UNets here). Cross-architecture merges match no weight keys and silently hand you model1 back.
The block names are the SD1.5 layout: time_embed. and label_emb. for the timestep and prompt conditioning, input_blocks.0.–input_blocks.11. for the encoder, middle_block.0.–middle_block.2. for the bottleneck, output_blocks.0.–output_blocks.11. for the decoder, and out. for the final conv.
There's no deep lore to add here, because nobody accumulated any. If you're merging SD2, you're a rare breed, and the node is there mainly so ComfyUI can claim full coverage of the family. Use it, but know that the effort is better spent on an SDXL or Flux merge - that's where the ecosystem actually lives. Ships with ComfyUI core; the SD1/SD2 merge node landed April 2024.
Inputs (32)
| Name | Type | Default | Description |
|---|---|---|---|
| model1 | MODEL | — | |
| model2 | MODEL | — | |
| time_embed. | FLOAT | 1.000–1 | — |
| label_emb. | FLOAT | 1.000–1 | — |
| input_blocks.0. | FLOAT | 1.000–1 | — |
| input_blocks.1. | FLOAT | 1.000–1 | — |
| input_blocks.2. | FLOAT | 1.000–1 | — |
| input_blocks.3. | FLOAT | 1.000–1 | — |
| input_blocks.4. | FLOAT | 1.000–1 | — |
| input_blocks.5. | FLOAT | 1.000–1 | — |
| input_blocks.6. | FLOAT | 1.000–1 | — |
| input_blocks.7. | FLOAT | 1.000–1 | — |
| input_blocks.8. | FLOAT | 1.000–1 | — |
| input_blocks.9. | FLOAT | 1.000–1 | — |
| input_blocks.10. | FLOAT | 1.000–1 | — |
| input_blocks.11. | FLOAT | 1.000–1 | — |
| middle_block.0. | FLOAT | 1.000–1 | — |
| middle_block.1. | FLOAT | 1.000–1 | — |
| middle_block.2. | FLOAT | 1.000–1 | — |
| output_blocks.0. | FLOAT | 1.000–1 | — |
| output_blocks.1. | FLOAT | 1.000–1 | — |
| output_blocks.2. | FLOAT | 1.000–1 | — |
| output_blocks.3. | FLOAT | 1.000–1 | — |
| output_blocks.4. | FLOAT | 1.000–1 | — |
| output_blocks.5. | FLOAT | 1.000–1 | — |
| output_blocks.6. | FLOAT | 1.000–1 | — |
| output_blocks.7. | FLOAT | 1.000–1 | — |
| output_blocks.8. | FLOAT | 1.000–1 | — |
| output_blocks.9. | FLOAT | 1.000–1 | — |
| output_blocks.10. | FLOAT | 1.000–1 | — |
| output_blocks.11. | FLOAT | 1.000–1 | — |
| out. | FLOAT | 1.000–1 | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |