Update app.py

app.py

@@ -4,44 +4,60 @@ import numpy as np
 import PIL.Image
 from PIL import Image
 import random
-from diffusers import ControlNetModel, StableDiffusionXLPipeline, AutoencoderKL
-from diffusers import DDIMScheduler, EulerAncestralDiscreteScheduler
-import cv2
+from diffusers import StableDiffusionXLPipeline
+from diffusers import EulerAncestralDiscreteScheduler
 import torch
 
 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
+# Make sure to use torch.float16 consistently throughout the pipeline
 pipe = StableDiffusionXLPipeline.from_pretrained(
     "votepurchase/waiNSFWIllustrious_v110",
     torch_dtype=torch.float16,
+    variant="fp16",  # Explicitly use fp16 variant
+    use_safetensors=True  # Use safetensors if available
 )
 
 pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
 pipe.to(device)
 
+# Force all components to use the same dtype
+pipe.text_encoder.to(torch.float16)
+pipe.text_encoder_2.to(torch.float16)
+pipe.vae.to(torch.float16)
+pipe.unet.to(torch.float16)
+
 MAX_SEED = np.iinfo(np.int32).max
 MAX_IMAGE_SIZE = 1216
-
     
 @spaces.GPU
 def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps):
-
+    # Check and truncate prompt if too long (CLIP can only handle 77 tokens)
+    if len(prompt.split()) > 60:  # Rough estimate to avoid exceeding token limit
+        print("Warning: Prompt may be too long and will be truncated by the model")
+        
     if randomize_seed:
         seed = random.randint(0, MAX_SEED)
 
-    generator = torch.Generator().manual_seed(seed)
-
-    output_image = pipe(
-        prompt=prompt,
-        negative_prompt=negative_prompt,
-        guidance_scale=guidance_scale,
-        num_inference_steps=num_inference_steps,
-        width=width,
-        height=height,
-        generator=generator
-    ).images[0]
-
-    return output_image
+    generator = torch.Generator(device=device).manual_seed(seed)
+    
+    try:
+        output_image = pipe(
+            prompt=prompt,
+            negative_prompt=negative_prompt,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            width=width,
+            height=height,
+            generator=generator
+        ).images[0]
+        
+        return output_image
+    except RuntimeError as e:
+        print(f"Error during generation: {e}")
+        # Return a blank image with error message
+        error_img = Image.new('RGB', (width, height), color=(0, 0, 0))
+        return error_img
 
 
 css = """
@@ -60,7 +76,7 @@ with gr.Blocks(css=css) as demo:
                 label="Prompt",
                 show_label=False,
                 max_lines=1,
-                placeholder="Enter your prompt",
+                placeholder="Enter your prompt (keep it under 60 words for best results)",
                 container=False,
             )
 
@@ -93,7 +109,7 @@ with gr.Blocks(css=css) as demo:
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=1024,#832,
+                    value=1024,
                 )
 
                 height = gr.Slider(
@@ -101,7 +117,7 @@ with gr.Blocks(css=css) as demo:
                     minimum=256,
                     maximum=MAX_IMAGE_SIZE,
                     step=32,
-                    value=1024,#1216,
+                    value=1024,
                 )
 
             with gr.Row():
@@ -121,7 +137,7 @@ with gr.Blocks(css=css) as demo:
                     value=28,
                 )
 
-    run_button.click(#lambda x: None, inputs=None, outputs=result).then(
+    run_button.click(
         fn=infer,
         inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
         outputs=[result]