Continuation of #1035 (#1120)

* remove batch size from repeat

* repeat empty string if uncond_tokens is none

* fix inpaint pipes

* return back whitespace to pass code quality

* Apply suggestions from code review

* Fix typos.
Co-authored-by: Had <had-95@yandex.ru>

examples/community/interpolate_stable_diffusion.py

@@ -278,7 +278,7 @@ class StableDiffusionWalkPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -307,7 +307,7 @@ class StableDiffusionWalkPipeline(DiffusionPipeline):
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
             seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
             uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.

examples/community/speech_to_image_diffusion.py

@@ -148,7 +148,7 @@ class SpeechToImagePipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -177,7 +177,7 @@ class SpeechToImagePipeline(DiffusionPipeline):
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
             seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
             uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.

examples/community/wildcard_stable_diffusion.py

@@ -295,7 +295,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -324,7 +324,7 @@ class WildcardStableDiffusionPipeline(DiffusionPipeline):
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
             seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
             uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -297,7 +297,7 @@ class StableDiffusionPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -326,7 +326,7 @@ class StableDiffusionPipeline(DiffusionPipeline):
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
             seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
             uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -295,7 +295,7 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -319,7 +319,9 @@ class StableDiffusionImg2ImgPipeline(DiffusionPipeline):
             uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
 
             # duplicate unconditional embeddings for each generation per prompt
-            uncond_embeddings = uncond_embeddings.repeat_interleave(batch_size * num_images_per_prompt, dim=0)
+            seq_len = uncond_embeddings.shape[1]
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.
             # Here we concatenate the unconditional and text embeddings into a single batch

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -302,7 +302,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -331,7 +331,7 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
 
             # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
             seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(batch_size, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
             uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py

@@ -284,7 +284,7 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
         if do_classifier_free_guidance:
             uncond_tokens: List[str]
             if negative_prompt is None:
-                uncond_tokens = [""]
+                uncond_tokens = [""] * batch_size
             elif type(prompt) is not type(negative_prompt):
                 raise TypeError(
                     f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
@@ -312,7 +312,9 @@ class StableDiffusionInpaintPipelineLegacy(DiffusionPipeline):
             uncond_embeddings = self.text_encoder(uncond_input.input_ids.to(self.device))[0]
 
             # duplicate unconditional embeddings for each generation per prompt
-            uncond_embeddings = uncond_embeddings.repeat_interleave(batch_size * num_images_per_prompt, dim=0)
+            seq_len = uncond_embeddings.shape[1]
+            uncond_embeddings = uncond_embeddings.repeat(1, num_images_per_prompt, 1)
+            uncond_embeddings = uncond_embeddings.view(batch_size * num_images_per_prompt, seq_len, -1)
 
             # For classifier free guidance, we need to do two forward passes.
             # Here we concatenate the unconditional and text embeddings into a single batch