The UI choice of SD3 nodes


For these three groups of nodes, I have the following improvement suggestions:

1. Since we already have a dedicated file `nodes_sd3.py`, I think we should combine `Load Checkpoint` and `ModelSamplingSD3` into one node: `SD3 Load Checkpoint`. This way, there's no empty clip tail, and users will know that "shift 3.0" is the model's hyperparameter.

2. On `TripleCLIPLoader`, I appraciate that Comfy made the node automatically determine three CLIPs, so users don't need to worry about choosing the CLIPs in the wrong order. However, users might not realize this, so it would be better to indicate the names of the CLIPs, allowing users to choose CLIPs more confidently. (I only know the order doesn't matter because I've read the code.) It would be great to retain the functionality of order insensitivity, but users don't need to know that. And the node can be called `SD3 Load CLIPs`, to be consistent.

3. I also noticed the note "Resolution should be around 1 megapixel and width/height must be multiple of 64", so I think it would be better to just build this rule into the `EmptySD3LatentImage`, calculating valid SD3 resolutions for users (there aren't many choices), since this is the dedicated Empty Latent Image generator for SD3. And it can be called `SD3 Empty Latent Image`, to be consistent.

These are my suggestions. And I've implemented that as a set of custom nodes for now. I'm not sure if I should submit a pull request or not, since there're so many PRs in line.

Here is the screenshot of my suggested nodes:

And here is the link: https://github.com/liusida/ComfyUI-SD3-nodes