How does ComfyUI implement the Conditioning (Set Mask) node?

[Pauan]

I've spent hours looking through the source code, but I can't find the actual implementation, maybe I'm blind and just missing it.

The node itself just sets the "mask" cond, but I can't find where the "mask" cond is being applied.

I'm trying to figure out if ComfyUI is using the same implementation as the webui regional prompter and if not, what the differences are.

[gershu-ar]

Correct me if I'm wrong: you’re not missing it — the mask node itself only sets the “mask” cond. The actual application happens inside the inpainting pipeline, specifically in the VAEEncodeForInpaint node. That’s where the image and mask are encoded together into latents, and the sampler then generates only within the masked regions.

So ComfyUI is not using the same implementation as WebUI’s regional prompter. Regional prompter works by splitting prompts across attention regions, while ComfyUI inpainting uses explicit spatial masks combined with inpainting-trained models.

In short: mask cond in ComfyUI = latent encoding with mask, not prompt-based conditioning.

[Pauan]

The actual application happens inside the inpainting pipeline, specifically in the VAEEncodeForInpaint node.

Hmmm, that doesn't sound right. Even if I don't use the VAEEncodeForInpaint node the Conditioning (Set Mask) node still works fine.

Also, VAEEncodeForInpaint doesn't actually do the implementation either, it only sets the "noise_mask", the implementation is somewhere else.

I'm trying to find where in the code it's actually processing things like "mask", "noise_mask", etc.