Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Let the VAEDecodeAudio node decode nested audio. (#15211)

comfyanonymouse803f24ea090
diff --git a/comfy_extras/nodes_audio.py b/comfy_extras/nodes_audio.py
index 4ac5ced53..1c6e91fe4 100644
--- a/comfy_extras/nodes_audio.py
+++ b/comfy_extras/nodes_audio.py
@@ -96,10 +96,14 @@ class VAEEncodeAudio(IO.ComfyNode):
def vae_decode_audio(vae, samples, tile=None, overlap=None):
+ latent = samples["samples"]
+ if latent.is_nested:
+ latent = latent.unbind()[-1]
+
if tile is not None:
- audio = vae.decode_tiled(samples["samples"], tile_x=tile, tile_y=tile, overlap=overlap).movedim(-1, 1)
+ audio = vae.decode_tiled(latent, tile_x=tile, tile_y=tile, overlap=overlap).movedim(-1, 1)
else:
- audio = vae.decode(samples["samples"]).movedim(-1, 1)
+ audio = vae.decode(latent).movedim(-1, 1)
std = torch.std(audio, dim=[1, 2], keepdim=True) * 5.0
std[std < 1.0] = 1.0