A

pipeline.py

@@ -366,7 +366,7 @@ class Zero123PlusPipeline(diffusers.StableDiffusionPipeline):
         global_embeds = encoded.image_embeds
         global_embeds = global_embeds.unsqueeze(-2)
         
-        if prompt_embeds is not None:
+        if prompt_embeds is None:
             encoder_hidden_states = self._encode_prompt(
                 prompts,
                 self.device,
@@ -375,6 +375,8 @@ class Zero123PlusPipeline(diffusers.StableDiffusionPipeline):
             )
             ramp = global_embeds.new_tensor(self.config.ramping_coefficients).unsqueeze(-1)
             encoder_hidden_states = encoder_hidden_states + global_embeds * ramp
+        else:
+            encoder_hidden_states = prompt_embeds
         cak = dict(cond_lat=cond_lat)
         if hasattr(self.unet, "controlnet"):
             cak['control_depth'] = depth_image