Update app.py

app.py

@@ -118,8 +118,10 @@ def infer_30(
     torch.set_float32_matmul_precision("highest")
     seed = random.randint(0, MAX_SEED)
     generator = torch.Generator(device='cuda').manual_seed(seed)
+    input_ids = pipe.tokenizer(prompt, return_tensors="pt").input_ids.to(device)
     print('-- generating image --')
     sd_image = pipe(
+            prompt_embeds = pipe.text_encoder(input_ids)[0], #ensure that the input_ids are on the correct device.
             prompt=prompt,
             prompt_2=prompt,
             prompt_3=prompt,
@@ -168,8 +170,10 @@ def infer_60(
     torch.set_float32_matmul_precision("highest")
     seed = random.randint(0, MAX_SEED)
     generator = torch.Generator(device='cuda').manual_seed(seed)
+    input_ids = pipe.tokenizer(prompt, return_tensors="pt").input_ids.to(device)
     print('-- generating image --')
     sd_image = pipe(
+            prompt_embeds = pipe.text_encoder(input_ids)[0], #ensure that the input_ids are on the correct device.
             prompt=prompt,
             prompt_2=prompt,
             prompt_3=prompt,
@@ -180,6 +184,7 @@ def infer_60(
             num_inference_steps=num_inference_steps,
             width=width,
             height=height,
+         #   cross_attention_kwargs={"scale": 0.75},
             generator=generator,
             max_sequence_length=512
     ).images[0]
@@ -217,8 +222,10 @@ def infer_90(
     torch.set_float32_matmul_precision("highest")
     seed = random.randint(0, MAX_SEED)
     generator = torch.Generator(device='cuda').manual_seed(seed)
+    input_ids = pipe.tokenizer(prompt, return_tensors="pt").input_ids.to(device)
     print('-- generating image --')
     sd_image = pipe(
+            prompt_embeds = pipe.text_encoder(input_ids)[0], #ensure that the input_ids are on the correct device.
             prompt=prompt,
             prompt_2=prompt,
             prompt_3=prompt,
@@ -229,6 +236,7 @@ def infer_90(
             num_inference_steps=num_inference_steps,
             width=width,
             height=height,
+         #   cross_attention_kwargs={"scale": 0.75},
             generator=generator,
             max_sequence_length=512
     ).images[0]