tiny fixes

lib/pipline_ConsistentID.py

@@ -362,7 +362,6 @@ class ConsistentIDPipeline(StableDiffusionPipeline):
             unet = edict()
             # Only keep the config and in_channels attributes that are used in the pipeline.
             unet.config = self.unet.config
-            unet.in_channels = self.unet.in_channels
             self.unet = unet
 
         if "vae" in released_components:
@@ -484,8 +483,6 @@ class ConsistentIDPipeline(StableDiffusionPipeline):
             prompt_embeds,
             negative_prompt_embeds,
         )
-        if not isinstance(input_subj_image_objs, list):
-            input_subj_image_objs = [input_subj_image_objs]
 
         # 2. Define call parameters
         if prompt is not None and isinstance(prompt, str):
@@ -499,22 +496,31 @@ class ConsistentIDPipeline(StableDiffusionPipeline):
         do_classifier_free_guidance = guidance_scale >= 1.0
         assert do_classifier_free_guidance
 
-        # 3. Encode input prompt
-        coarse_prompt_embeds, fine_prompt_embeds = \
-            self.extract_double_id_prompt_embeds(prompt, negative_prompt, input_subj_image_objs[0], device)
+        if input_subj_image_objs is not None:
+            if not isinstance(input_subj_image_objs, list):
+                input_subj_image_objs = [input_subj_image_objs]
+
+            # 3. Encode input prompt
+            coarse_prompt_embeds, fine_prompt_embeds = \
+                self.extract_double_id_prompt_embeds(prompt, negative_prompt, input_subj_image_objs[0], device)
+        else:
+            # Hijack the coarse_prompt_embeds and fine_prompt_embeds to be the input prompt_embeds.
+            cfg_prompt_embeds    = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+            coarse_prompt_embeds = cfg_prompt_embeds
+            fine_prompt_embeds   = cfg_prompt_embeds
 
         # 7. Prepare timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)
         timesteps = self.scheduler.timesteps
 
         # 8. Prepare latent variables
-        num_channels_latents = self.unet.in_channels
+        num_channels_latents = self.unet.config.in_channels
         latents = self.prepare_latents(
             batch_size * num_images_per_prompt,
             num_channels_latents,
             height,
             width,
-            coarse_prompt_embeds.dtype,
+            self.dtype,
             device,
             generator,
             latents,