nodes/hum-ma-ComfyUI-TiledVaeLite
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add files via upload

Browse Source 
Fix errors when decoding images with last_frame_fix enabled
This commit is contained in:
hum_ma
2025-12-08 01:45:51 +02:00
committed by GitHub
parent 6be81dd946
commit 583ca4dd39
1 changed files with 9 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tiled_vae_decode.py
+9 -10
View File
@@ -38,16 +38,15 @@ class LTTiledVAEDecode:
# Get the latent samples
samples = latents["samples"]
if last_frame_fix:
# Repeat the last frame along dimension 2 (frames)
# samples: [batch, channels, frames, height, width]
last_frame = samples[
:, :, -1:, :, :
] # shape: [batch, channels, 1, height, width]
samples = torch.cat([samples, last_frame], dim=2)
if samples.ndim == 5: # video latent
if last_frame_fix:
# Repeat the last frame along dimension 2 (frames)
# samples: [batch, channels, frames, height, width]
last_frame = samples[
:, :, -1:, :, :
] # shape: [batch, channels, 1, height, width]
samples = torch.cat([samples, last_frame], dim=2)
batch, channels, frames, height, width = samples.shape
time_scale_factor, width_scale_factor, height_scale_factor = (
vae.downscale_index_formula
@@ -211,7 +210,7 @@ class LTTiledVAEDecode:
batch * image_frames, output_height, output_width, output.shape[-1]
)
if last_frame_fix:
if last_frame_fix and samples.ndim == 5:
output = output[:-time_scale_factor, :, :]
time_total = time.perf_counter()-time_init