The audio half of 😺NKD AV Latent on its own, for when you'd rather keep the chain out in the open. It is the audio branch's Set Latent Noise Mask, latent in and the same latent out with the mask already on it, so it replaces both that node and the Solid Mask feeding it, and the audio latent gets wired once instead of forked.
flowchart LR
VE(["VAE Encode Audio"]):::external --> AM
MK(["Mask"]):::input --> AM
AM["**NKD Audio Mask**"]:::nkd -- audio_latent --> CC(["Concat AV Latent"]):::external
classDef nkd fill:#3b3b6b,stroke:#8ab4ff,stroke-width:2px,color:#fff
classDef input fill:#2d2d2d,stroke:#888,color:#eee
classDef external fill:#2d2d2d,stroke:#888,color:#eee
Same retiming as follow mask, and it arrives already fitted to the soundtrack, so nothing stretches it on the way into the sampler. Different models lay their audio out differently, so the timing is read off the latent you connect rather than assumed, and a latent it can't read gets refused with a message instead of quietly masking the wrong thing. Tested against MiniMax H3; LTXV should work but hasn't been run on a real clip.
It isn't a shortcut for the stock node. Hand a video mask to Set Latent Noise Mask and the soundtrack comes back regenerated from end to end no matter what the mask said, with no error to tell you. It reads a mask as a picture, and a soundtrack has no picture in it.