Merge pull request #39 from ModelTC/wangshankun-wan_target_video_len_hard_code

Fix  bug: wan's target_video_len hard code

lightx2v/models/runners/wan/wan_runner.py

@@ -113,7 +113,7 @@ class WanRunner(DefaultRunner):
                 torch.concat(
                     [
                         torch.nn.functional.interpolate(img[None].cpu(), size=(h, w), mode="bicubic").transpose(0, 1),
-                        torch.zeros(3, 80, h, w),
+                        torch.zeros(3, config.target_video_length - 1, h, w),
                     ],
                     dim=1,
                 ).cuda()