small fixes

hyvideo/diffusion/pipelines/pipeline_hunyuan_video.py

@@ -451,7 +451,12 @@ class HunyuanVideoPipeline(DiffusionPipeline):
             if prompt_mask is not None:
                 prompt_mask = torch.cat([prompt_mask, prompt_mask])
             if prompt_embeds_2 is not None:
-                prompt_embeds_2 = torch.cat([prompt_embeds_2, prompt_embeds_2])            
+                prompt_embeds_2 = torch.cat([prompt_embeds_2, prompt_embeds_2])
+
+        prompt_embeds = prompt_embeds.to(device = device, dtype = self.base_dtype)
+        prompt_mask = prompt_mask.to(device)
+        if prompt_embeds_2 is not None:
+            prompt_embeds_2 = prompt_embeds_2.to(device = device, dtype = self.base_dtype)
 
         # 4. Prepare timesteps
         extra_set_timesteps_kwargs = self.prepare_extra_func_kwargs(
@@ -564,7 +569,6 @@ class HunyuanVideoPipeline(DiffusionPipeline):
                 else:
                     guidance_expand = None
                 
-                #print("latent_model_input", latent_model_input.shape, "guidance_expand", guidance_expand)
                 # predict the noise residual
                 with torch.autocast(
                     device_type="cuda", dtype=self.base_dtype, enabled=True

hyvideo/modules/models.py

@@ -665,7 +665,7 @@ class HYVideoDiffusionTransformer(ModelMixin, ConfigMixin):
     ) -> Union[torch.Tensor, Dict[str, torch.Tensor]]:
         out = {}
         img = x
-        txt = text_states.to(x.device)
+        txt = text_states
         _, _, ot, oh, ow = x.shape
         tt, th, tw = (
             ot // self.patch_size[0],
@@ -678,7 +678,7 @@ class HYVideoDiffusionTransformer(ModelMixin, ConfigMixin):
 
         # text modulation
         if text_states_2 is not None:
-            vec = vec + self.vector_in(text_states_2.to(x.device))
+            vec = vec + self.vector_in(text_states_2)
 
         # guidance modulation
         if guidance is not None:
@@ -695,7 +695,7 @@ class HYVideoDiffusionTransformer(ModelMixin, ConfigMixin):
         if self.text_projection == "linear":
             txt = self.txt_in(txt)
         elif self.text_projection == "single_refiner":
-            txt = self.txt_in(txt, t, text_mask.to(x.device) if self.use_attention_mask else None)
+            txt = self.txt_in(txt, t, text_mask if self.use_attention_mask else None)
         else:
             raise NotImplementedError(
                 f"Unsupported text_projection: {self.text_projection}"

nodes.py

@@ -848,6 +848,8 @@ class HyVideoTextEncode:
                 tokens = clip_l.tokenize(negative_prompt, return_word_ids=True)
                 negative_prompt_embeds_2 = clip_l.encode_from_tokens(tokens, return_pooled=True, return_dict=False)[1]
                 negative_prompt_embeds_2 = negative_prompt_embeds_2.to(device=device)
+            else:
+                negative_prompt_embeds_2 = None
             attention_mask_2, negative_attention_mask_2 = None, None
 
             if force_offload: