zlei6/ComfyUI
1
0
Fork 0
mirror of https://github.com/Comfy-Org/ComfyUI.git synced 2026-08-25 10:22:12 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix VAEDecodeTiled crash on NestedTensor latents (MiniMax H3) (#15477)

Browse Source 
VAEDecode unwraps a NestedTensor latent (video/audio pair) to its
video component before calling vae.decode(). VAEDecodeTiled skipped
this unwrap and passed the NestedTensor straight into
vae.decode_tiled(), which fails deep in the MiniMax H3 video VAE when
a real tensor's .to() is called with the NestedTensor as an argument.

Fixes #15468.
This commit is contained in:
chelsealong
2026-08-11 11:00:23 +08:00
committed by GitHub
parent 34744cd29e
commit 6233790c6d
2 changed files with 32 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
nodes.py
View File

@@ -364,8 +364,12 @@ class VAEDecodeTiled:
temporal_size = None
temporal_overlap = None
latent = samples["samples"]
if latent.is_nested:
latent = latent.unbind()[0]
compression = vae.spacial_compression_decode()
images = vae.decode_tiled(samples["samples"], tile_x=tile_size // compression, tile_y=tile_size // compression, overlap=overlap // compression, tile_t=temporal_size, overlap_t=temporal_overlap)
images = vae.decode_tiled(latent, tile_x=tile_size // compression, tile_y=tile_size // compression, overlap=overlap // compression, tile_t=temporal_size, overlap_t=temporal_overlap)
if len(images.shape) == 5: #Combine batches
images = images.reshape(-1, images.shape[-3], images.shape[-2], images.shape[-1])
return (images, )