diff --git a/opensora/models/diffusion/latte/modeling_latte.py b/opensora/models/diffusion/latte/modeling_latte.py
index fa8ed7192..0baadb3b0 100644
--- a/opensora/models/diffusion/latte/modeling_latte.py
+++ b/opensora/models/diffusion/latte/modeling_latte.py
@@ -401,7 +401,7 @@ class LatteT2V(ModelMixin, ConfigMixin):
                 encoder_hidden_states = torch.cat([encoder_hidden_states_video, encoder_hidden_states_image], dim=1)
                 encoder_hidden_states_spatial = rearrange(encoder_hidden_states, 'b f t d -> (b f) t d').contiguous()
             else:
-                encoder_hidden_states_spatial = repeat(encoder_hidden_states, 'b t d -> (b f) t d', f=frame).contiguous()
+                encoder_hidden_states_spatial = repeat(encoder_hidden_states, 'b 1 t d -> (b f) t d', f=frame).contiguous()
 
         # prepare timesteps for spatial and temporal block
         timestep_spatial = repeat(timestep, 'b d -> (b f) d', f=frame + use_image_num).contiguous()
@@ -673,4 +673,4 @@ if __name__ == '__main__':
                         encoder_attention_mask=cond_mask, use_image_num=args.use_image_num, timestep=timestep)
     with torch.no_grad():
         output = model(**model_kwargs)
-    # print(output)
\ No newline at end of file
+    # print(output)