Fix clip_skip AttributeError on Stable Diffusion pipelines with transformers>=5.6

src/diffusers/pipelines/animatediff/pipeline_animatediff.py

@@ -256,7 +256,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/animatediff/pipeline_animatediff_controlnet.py

@@ -301,7 +301,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/animatediff/pipeline_animatediff_sparsectrl.py

@@ -310,7 +310,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py

@@ -358,7 +358,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/animatediff/pipeline_animatediff_video2video_controlnet.py

@@ -389,7 +389,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/controlnet/pipeline_controlnet.py

@@ -400,7 +400,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py

@@ -378,7 +378,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py

@@ -384,7 +384,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion.py

@@ -429,7 +429,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion_img2img.py

@@ -457,7 +457,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/controlnet_xs/pipeline_controlnet_xs.py

@@ -334,7 +334,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/i2vgen_xl/pipeline_i2vgen_xl.py

@@ -245,7 +245,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype
@@ -315,7 +319,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                negative_prompt_embeds = self.text_encoder.text_model.final_layer_norm(negative_prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                negative_prompt_embeds = text_model.final_layer_norm(negative_prompt_embeds)
 
         if self.do_classifier_free_guidance:
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method

src/diffusers/pipelines/deprecated/pia/pipeline_pia.py

@@ -318,7 +318,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_attend_and_excite/pipeline_stable_diffusion_attend_and_excite.py

@@ -398,7 +398,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_diffedit/pipeline_stable_diffusion_diffedit.py

@@ -524,7 +524,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_gligen/pipeline_stable_diffusion_gligen.py

@@ -322,7 +322,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_gligen/pipeline_stable_diffusion_gligen_text_image.py

@@ -353,7 +353,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py

@@ -414,7 +414,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

@@ -385,7 +385,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_sag/pipeline_stable_diffusion_sag.py

@@ -313,7 +313,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_cycle_diffusion.py

@@ -390,7 +390,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_inpaint_legacy.py

@@ -361,7 +361,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_model_editing.py

@@ -294,7 +294,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype

src/diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_paradigms.py

@@ -291,7 +291,11 @@ def encode_prompt(
                 # representations. The `last_hidden_states` that we typically use for
                 # obtaining the final prompt representations passes through the LayerNorm
                 # layer.
-                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+                # CLIPTextModel was flattened in transformers>=5.6 (no longer wrapped in .text_model).
+                text_model = (
+                    self.text_encoder.text_model if hasattr(self.text_encoder, "text_model") else self.text_encoder
+                )
+                prompt_embeds = text_model.final_layer_norm(prompt_embeds)
 
         if self.text_encoder is not None:
             prompt_embeds_dtype = self.text_encoder.dtype