Wan2.2-TI2V VAE encode or decode Bug #1305
Description
HI ！我在运行examples/wanvideo/model_inference/Wan2.2-TI2V-5B.py时，在VAE decode阶段遇到报错：
File "DiffSynth-Studio/diffsynth/pipelines/wan_video.py", line 328, in call
video = self.vae.decode(inputs_shared["latents"], device=self.device, tiled=tiled, tile_size=tile_size, tile_stride=tile_stride)
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 1241, in decode
video = self.tiled_decode(hidden_state, device, tile_size, tile_stride)
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 1126, in tiled_decode
hidden_states_batch = self.model.decode(hidden_states_batch, self.scale).to(data_device)
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 1340, in decode
out, self._feat_map, self._conv_idx = self.decoder(x[:, :, i:i + 1, :, :],
……
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 509, in forward
x_main = module(x_main, feat_cache, feat_idx)
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 300, in forward
x = layer(x)
File "DiffSynth-Studio/diffsynth/models/wan_video_vae.py", line 68, in forward
return F.normalize(
AttributeError: 'tuple' object has no attribute 'norm'
并且1、当我使用我自己自定义的pipeline（基于Wan2.2-TI2V-5B），修改了模型的in dim后，encode阶段也会出现这个报错；
2、运行Wan2.1的pipeline的时候并未遇到这个报错。