search

comfyanonymous / ComfyUI

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface.
https://www.comfy.org/
GNU General Public License v3.0
56.32k stars 5.97k forks source link

Cleanup of latent image size picker nodes #5361

Closed pandayummy closed 2 weeks ago

pandayummy commented 2 weeks ago

Feature Idea

We have 3 nodes of the same function in the core now, can we have a cleanup?

image

Existing Solutions

No response

Other

No response

ltdrdata commented 2 weeks ago

EmptySD3LatentImage can be replaced with Empty Latent. However, strictly speaking, Empty Latent is a 4-channel latent used in SD1/SDXL, while EmptySD3LatentImage is a 16-channel latent, so they cannot be mixed. It's just that, for user convenience, when using 16-channel models (SD3, FLUX) in KSampler, a function is provided to convert 4 channels to 16 channels, but this only applies to empty latents.

On the other hand, StableCascade_EmptyLatentImage can never be integrated. Stable Cascade uses different types of latents for each stage, and because it has a very heterogeneous structure that requires some form of connection between those latents, it can never be compatible with other models.

If it's simply a matter of selecting resolution, you can change each widget to an input and connect a size picker custom node.