From d1b69c90593a55926607995f6b07826a873497f6 Mon Sep 17 00:00:00 2001
From: yangxy57 <1524552292@qq.com>
Date: Fri, 13 Feb 2026 11:28:37 +0800
Subject: [PATCH] Fix missing .unsqueeze(1) in attention mask for HunyuanVideo
 VAE

---
 .../models/autoencoders/autoencoder_kl_hunyuanvideo15.py        | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py b/src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py
index c662c1657513..dea370c65c58 100644
--- a/src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py
+++ b/src/diffusers/models/autoencoders/autoencoder_kl_hunyuanvideo15.py
@@ -145,7 +145,7 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
             frames, height * width, query.dtype, query.device, batch_size=batch_size
         )
 
-        x = nn.functional.scaled_dot_product_attention(query, key, value, attn_mask=attention_mask)
+        x = nn.functional.scaled_dot_product_attention(query, key, value, attn_mask=attention_mask.unsqueeze(1))
 
         # batch_size, 1, frames * height * width, channels