ComfyUI-LTXAudioLatentTrim

LatentStripMask

Removes the noisemask key from a latent dict. Why it exists: LTXVAddLatents attempts to merge noise masks from both inputs. When a trimmed audio latent still carries its mask from a previous generation pass, the merge logic fails with a dimension mismatch. Stripping the mask before concatenation avoids this.

Latent · 1 Eingabe · 0 Parameter · 1 Ausgabe

LTXAudioLatentTrim

Trims an LTX audio latent [B, C, T, F] along the temporal dimension using start/end latent frame indices. Supports negative indexing (endindex=-1 = last frame inclusive). Why it exists: LTX2 audio latents are 4D tensors [B, C, T, F], while video latents are 5D [B, C, T, H, W]. The existing trimming nodes in ComfyUI and KJNodes (LTXVSelectLatents, GetLatentRangeFromBatch) either expect 5D tensors or index the batch dimension rather than the temporal dimension — making them unusable for audio latent trimming in sliding window workflows. This node handles the 4D case correctly.

Latent/Audio · 1 Eingabe · 3 Parameter · 1 Ausgabe