update

musetalk/pipeline.py

@@ -109,8 +109,8 @@ def apply_musetalk_lipsync(
         crop_frame = crop_frame.to(vae.device, dtype=vae.dtype)
 
         with torch.no_grad():
-            masked_latents = vae.encode(crop_frame).sample
-            latents = vae.encode(crop_frame).sample
+            masked_latents = vae.encode(crop_frame).latent_dist.sample()
+            latents = vae.encode(crop_frame).latent_dist.sample()
             latent_model_input = torch.cat([masked_latents, latents], dim=1)
             input_latent_list.append(latent_model_input)
     print(f"✓ Encoded {len(input_latent_list)} frames")