update demo

app.py

@@ -33,8 +33,8 @@ def infer(prompt: str, progress=gr.Progress(track_tqdm=True)):
             negative_prompt="",
             num_inference_steps=50,
             guidance_scale=7.5,
-            width=432,
-            height=240, #480x288  624x352 432x240 768x432
+            width=768,
+            height=432, #480x288  624x352 432x240 768x432
             frames=16
         )
     

models/attention.py

@@ -1362,7 +1362,7 @@ class VchitectAttnProcessor:
         cross_output = rearrange(cross_output, "B (S T) C -> (B T) S C", T=Frame, B=batchsize)
         cross_output = attn.to_out_context(cross_output)
 
-        hidden_states = hidden_states*1.1 + cross_output
+        hidden_states = hidden_states + cross_output
 
         # Split the attention outputs.
         hidden_states, encoder_hidden_states = (