provide empty negative prompt when training

StableDiffuser.py

@@ -114,9 +114,11 @@ class StableDiffuser(torch.nn.Module):
         latents = noise * self.scheduler.init_noise_sigma
         return latents
 
-    def get_text_embeddings(self, prompts, negative_prompts, n_imgs):
+    def get_text_embeddings(self, prompts, negative_prompts=None, n_imgs=1):
         text_tokens = self.text_tokenize(prompts)
         text_embeddings = self.text_encode(text_tokens)
+        if negative_prompts is None:
+            negative_prompts = [""] * len(prompts)
         unconditional_tokens = self.text_tokenize(negative_prompts)
         unconditional_embeddings = self.text_encode(unconditional_tokens)
         text_embeddings = torch.cat([unconditional_embeddings, text_embeddings]).repeat_interleave(n_imgs, dim=0)

train.py

@@ -36,11 +36,9 @@ def train(repo_id_or_path, img_size, prompt, modules, freeze_modules, iterations
             optimizer.zero_grad()
 
             iteration = torch.randint(1, nsteps - 1, (1,)).item()
-
             latents = diffuser.get_initial_latents(1, img_size, 1)
 
             with finetuner:
-
                 latents_steps, _ = diffuser.diffusion(
                     latents,
                     positive_text_embeddings,