|
1 | | -The CheckpointSave node is designed for saving the state of various model components, including models, CLIP, and VAE, into a checkpoint file. This functionality is crucial for preserving the training progress or configuration of models for later use or sharing. |
| 1 | +The `Save Checkpoint` node is designed to save a complete Stable Diffusion model (including UNet, CLIP, and VAE components) as a **.safetensors** format checkpoint file. |
| 2 | + |
| 3 | +The Save Checkpoint is primarily used in model merging workflows. After creating a new merged model through nodes like `ModelMergeSimple`, `ModelMergeBlocks`, etc., you can use this node to save the result as a reusable checkpoint file. |
2 | 4 |
|
3 | 5 | ## Inputs |
4 | 6 |
|
5 | | -| Parameter | Data Type | Description | |
6 | | -|-----------|-------------|-------------| |
7 | | -| `model` | MODEL | The model parameter represents the primary model whose state is to be saved. It is essential for capturing the current state of the model for future restoration or analysis. | |
8 | | -| `clip` | CLIP | The clip parameter is intended for the CLIP model associated with the primary model, allowing its state to be saved alongside the main model. | |
9 | | -| `vae` | VAE | The vae parameter is for the Variational Autoencoder (VAE) model, enabling its state to be saved for future use or analysis alongside the main model and CLIP. | |
10 | | -| `filename_prefix` | STRING | This parameter specifies the prefix for the filename under which the checkpoint will be saved, providing a means to organize and identify saved checkpoints. | |
| 7 | +| Parameter | Data Type | Description | |
| 8 | +|-----------|-----------|-------------| |
| 9 | +| `model` | MODEL | The model parameter represents the primary model whose state is to be saved. It is essential for capturing the current state of the model for future restoration or analysis. | |
| 10 | +| `clip` | CLIP | The clip parameter is intended for the CLIP model associated with the primary model, allowing its state to be saved alongside the main model. | |
| 11 | +| `vae` | VAE | The vae parameter is for the Variational Autoencoder (VAE) model, enabling its state to be saved for future use or analysis alongside the main model and CLIP. | |
| 12 | +| `filename_prefix` | STRING | This parameter specifies the prefix for the filename under which the checkpoint will be saved. | |
| 13 | + |
| 14 | +Additionally, the node has two hidden inputs for metadata: |
| 15 | + |
| 16 | +**prompt (PROMPT)**: Workflow prompt information |
| 17 | +**extra_pnginfo (EXTRA_PNGINFO)**: Additional PNG information |
11 | 18 |
|
12 | 19 | ## Outputs |
13 | 20 |
|
14 | 21 | This node will output a checkpoint file, and the corresponding output file path is `output/checkpoints/` directory |
| 22 | + |
| 23 | +## Architecture Compatibility |
| 24 | + |
| 25 | +- Currently fully supported: SDXL, SD3, SVD and other mainstream architectures, see [source code](https://github.com/comfyanonymous/ComfyUI/blob/master/comfy_extras/nodes_model_merging.py#L176-L189) |
| 26 | +- Basic support: Other architectures can be saved but without standardized metadata information |
| 27 | + |
| 28 | +## Related Links |
| 29 | + |
| 30 | +Related source code: [nodes_model_merging.py#L227](https://github.com/comfyanonymous/ComfyUI/blob/master/comfy_extras/nodes_model_merging.py#L227) |
0 commit comments