fix image encoding

examples/opensora_hpcai/opensora/pipelines/infer_pipeline.py

@@ -7,6 +7,7 @@
 
 from mindone.models.modules.pos_embed import get_2d_sincos_pos_embed
 
+from ..models.vae.vae import VideoAutoencoderKL, VideoAutoencoderPipeline
 from ..models.layers.rotary_embedding import precompute_freqs_cis
 from ..schedulers.iddpm import create_diffusion
 
@@ -61,7 +62,12 @@ def __init__(
     # @ms.jit
     def vae_encode(self, x: Tensor) -> Tensor:
         # image_latents = ops.stop_gradient(self.vae.encode(x))
-        image_latents = ops.stop_gradient(self.vae.module.encode(x) * self.vae.scale_factor)
+        if isinstance(self.vae, VideoAutoencoderKL):
+            spatial_vae = self.vae
+        elif isinstance(self.vae, VideoAutoencoderPipeline):
+            spatial_vae = self.vae.spatial_vae
+        # TODO: unify scale inside vae class
+        image_latents = ops.stop_gradient(spatial_vae.module.encode(x) * spatial_vae.scale_factor)
         return image_latents
 
     def vae_decode(self, x: Tensor) -> Tensor: