Fix Cascade empty prompt encode

Author: Disty0

modules/prompt_parser_diffusers.py

@@ -253,6 +253,8 @@ def prepare_embedding_providers(pipe, clip_skip) -> list[EmbeddingsProvider]:
     if getattr(pipe, "prior_pipe", None) is not None and getattr(pipe.prior_pipe, "tokenizer", None) is not None and getattr(pipe.prior_pipe, "text_encoder", None) is not None:
         provider = EmbeddingsProvider(padding_attention_mask_value=0, tokenizer=pipe.prior_pipe.tokenizer, text_encoder=pipe.prior_pipe.text_encoder, truncate=False, returned_embeddings_type=embedding_type, device=device)
         embeddings_providers.append(provider)
+        no_mask_provider = EmbeddingsProvider(padding_attention_mask_value=1, tokenizer=pipe.prior_pipe.tokenizer, text_encoder=pipe.prior_pipe.text_encoder, truncate=False, returned_embeddings_type=embedding_type, device=device)
+        embeddings_providers.append(no_mask_provider)
     elif getattr(pipe, "tokenizer", None) is not None and getattr(pipe, "text_encoder", None) is not None:
         provider = EmbeddingsProvider(tokenizer=pipe.tokenizer, text_encoder=pipe.text_encoder, truncate=False, returned_embeddings_type=embedding_type, device=device)
         embeddings_providers.append(provider)
@@ -262,7 +264,7 @@ def prepare_embedding_providers(pipe, clip_skip) -> list[EmbeddingsProvider]:
     return embeddings_providers
 
 
-def pad_to_same_length(pipe, embeds):
+def pad_to_same_length(pipe, embeds, empty_embedding_providers=None):
     if not hasattr(pipe, 'encode_prompt') and 'StableCascade' not in pipe.__class__.__name__:
         return embeds
     device = pipe.device if str(pipe.device) != 'meta' else devices.device
@@ -271,8 +273,8 @@ def pad_to_same_length(pipe, embeds):
     else:
         try:
             if 'StableCascade' in pipe.__class__.__name__:
-                empty_embed = pipe.prior_pipe.encode_prompt(device, 1, 1, False, prompt="")
-                empty_embed = [torch.nn.functional.normalize(empty_embed[0])]
+                empty_embed = empty_embedding_providers[0].get_embeddings_for_weighted_prompt_fragments(text_batch=[[""]], fragment_weights_batch=[[1]], should_return_tokens=False, device=device)
+                empty_embed = [empty_embed]
             else:
                 empty_embed = pipe.encode_prompt("")
         except TypeError:  # SD1.5
@@ -331,6 +333,11 @@ def get_weighted_text_embeddings(pipe, prompt: str = "", neg_prompt: str = "", c
         negative_weights.pop(0)
 
     embedding_providers = prepare_embedding_providers(pipe, clip_skip)
+    empty_embedding_providers = None
+    if 'StableCascade' in pipe.__class__.__name__:
+        empty_embedding_providers = [embedding_providers[1]]
+        embedding_providers = [embedding_providers[0]]
+
     prompt_embeds = []
     negative_prompt_embeds = []
     pooled_prompt_embeds = []
@@ -400,7 +407,7 @@ def get_weighted_text_embeddings(pipe, prompt: str = "", neg_prompt: str = "", c
         negative_pooled_prompt_embeds = None
     debug(f'Prompt: positive={prompt_embeds.shape if prompt_embeds is not None else None} pooled={pooled_prompt_embeds.shape if pooled_prompt_embeds is not None else None} negative={negative_prompt_embeds.shape if negative_prompt_embeds is not None else None} pooled={negative_pooled_prompt_embeds.shape if negative_pooled_prompt_embeds is not None else None}')
     if prompt_embeds.shape[1] != negative_prompt_embeds.shape[1]:
-        [prompt_embeds, negative_prompt_embeds] = pad_to_same_length(pipe, [prompt_embeds, negative_prompt_embeds])
+        [prompt_embeds, negative_prompt_embeds] = pad_to_same_length(pipe, [prompt_embeds, negative_prompt_embeds], empty_embedding_providers=empty_embedding_providers)
     if SD3:
         device = pipe.device if str(pipe.device) != 'meta' else devices.device
         t5_prompt_embed = pipe._get_t5_prompt_embeds( # pylint: disable=protected-access