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.

Fix LTXAV crash when sampling without an audio latent (#15132)

Barish Ozbaye651b7bef55a
diff --git a/comfy/ldm/lightricks/model.py b/comfy/ldm/lightricks/model.py
index 92bb8118c..f9de3a38e 100644
--- a/comfy/ldm/lightricks/model.py
+++ b/comfy/ldm/lightricks/model.py
@@ -671,9 +671,9 @@ def freqs_cis_matrix(freqs, pad_size, split_mode, num_attention_heads, out_dtype
cos_freq = torch.cat((cos_padding, cos_freq), dim=-1)
sin_freq = torch.cat((sin_padding, sin_freq), dim=-1)
- B, T, _ = cos_freq.shape
- cos_freq = cos_freq.reshape(B, T, num_attention_heads, -1)
- sin_freq = sin_freq.reshape(B, T, num_attention_heads, -1)
+ B, T, half_HD = cos_freq.shape
+ cos_freq = cos_freq.reshape(B, T, num_attention_heads, half_HD // num_attention_heads)
+ sin_freq = sin_freq.reshape(B, T, num_attention_heads, half_HD // num_attention_heads)
rotation_matrix = torch.stack(
(cos_freq, -sin_freq, sin_freq, cos_freq), dim=-1
)