mps

app.py

@@ -11,14 +11,14 @@ import numpy as np
 from PIL import Image
 from datetime import datetime
 
-preferred_device = "cuda" if torch.cuda.is_available() else "cpu"
-preferred_dtype = torch.float32 if preferred_device == 'cpu' else torch.float16
+preferred_device = "cuda" if torch.cuda.is_available() else ("mps" if torch.backends.mps.is_available() else "cpu")
+preferred_dtype = torch.float16 if preferred_device == 'cuda' else torch.float32
 
 def label_func(fn): return path/"labels"/f"{fn.stem}_P{fn.suffix}"
 
 segmodel = load_learner("camvid-512.pkl")
 
-if preferred_device == "cuda":
+if preferred_dtype == torch.float16:
     segmodel = segmodel.to_fp16()
 
 inpainting_pipeline = StableDiffusionInpaintPipeline.from_pretrained(
@@ -29,7 +29,7 @@ inpainting_pipeline = StableDiffusionInpaintPipeline.from_pretrained(
 
 working_size = (512, 512)
 
-default_inpainting_prompt = "watercolor of a leafy pedestrian mall at golden hour with multiracial genderqueer joggers and bicyclists and wheelchair users talking and laughing"
+default_inpainting_prompt = "award-winning photo of a leafy pedestrian mall full of people, with multiracial genderqueer joggers and bicyclists and wheelchair users talking and laughing"
 
 seg_vocabulary = ['Animal', 'Archway', 'Bicyclist', 'Bridge', 'Building', 'Car',
        'CartLuggagePram', 'Child', 'Column_Pole', 'Fence', 'LaneMkgsDriv',
@@ -64,7 +64,7 @@ def app(img, prompt):
         image=img,
         mask_image=mask,
         strength=0.95,
-        num_inference_steps=50,
+        num_inference_steps=20,
     ).images[0]
     end_time = datetime.now().timestamp()
     draw = ImageDraw.Draw(overlay_img)