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.

LTXV VAE decoder: switch default padding mode (#8930)

Harel Cain9bc2798f7294
diff --git a/comfy/ldm/lightricks/vae/causal_video_autoencoder.py b/comfy/ldm/lightricks/vae/causal_video_autoencoder.py
index f91870d71..75ed069ad 100644
--- a/comfy/ldm/lightricks/vae/causal_video_autoencoder.py
+++ b/comfy/ldm/lightricks/vae/causal_video_autoencoder.py
@@ -973,7 +973,7 @@ class VideoVAE(nn.Module):
norm_layer=config.get("norm_layer", "group_norm"),
causal=config.get("causal_decoder", False),
timestep_conditioning=self.timestep_conditioning,
- spatial_padding_mode=config.get("spatial_padding_mode", "zeros"),
+ spatial_padding_mode=config.get("spatial_padding_mode", "reflect"),
)
self.per_channel_statistics = processor()