Test

pipeline.py

@@ -332,9 +332,7 @@ class Zero123PlusPipeline(diffusers.StableDiffusionPipeline):
         self,
         image: Image.Image = None,
         prompt = "",
-        negative_prompt = "",
         prompt_embeds = None,
-        negative_prompt_embeds = None,
         *args,
         num_images_per_prompt: Optional[int] = 1,
         guidance_scale=4.0,
@@ -368,19 +366,15 @@ class Zero123PlusPipeline(diffusers.StableDiffusionPipeline):
         global_embeds = encoded.image_embeds
         global_embeds = global_embeds.unsqueeze(-2)
         
-        if prompt_embeds is None and negative_prompt_embeds is None:
+        if prompt_embeds is None:
             prompt_embeds, negative_prompt_embeds = self.encode_prompt(
                 prompt,
                 self.device,
                 num_images_per_prompt,
                 False,
-                negative_prompt=negative_prompt,
             )
-            encoder_hidden_states = torch.cat([prompt_embeds, negative_prompt_embeds])
-            ramp = global_embeds.new_tensor(self.config.ramping_coefficients).unsqueeze(-1)
-            encoder_hidden_states = encoder_hidden_states + global_embeds * ramp
-        else:
-            encoder_hidden_states = torch.cat([prompt_embeds, negative_prompt_embeds])
+        ramp = global_embeds.new_tensor(self.config.ramping_coefficients).unsqueeze(-1)
+        encoder_hidden_states = torch.cat([prompt_embeds, torch.zeros(prompt_embeds.shape)]) + global_embeds * ramp
         cak = dict(cond_lat=cond_lat)
         if hasattr(self.unet, "controlnet"):
             cak['control_depth'] = depth_image