Update pipeline_cogvideox.py

This commit is contained in:
kijai 2024-12-20 17:00:42 +02:00
parent 0758d2d016
commit fcc0f3e65a

View File

@ -515,6 +515,7 @@ class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
image_cond_latents = torch.cat([first_frame, image_cond_latents], dim=1)
fun_mask_first_frame = fun_mask[:, : image_cond_frame_count % patch_size_t, ...]
fun_mask = torch.cat([fun_mask_first_frame, fun_mask], dim=1)
fun_mask[:, 1:, ...] = 0
image_cond_latents = image_cond_latents.to(self.vae_dtype)
# 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline