512

app.py

@@ -34,7 +34,8 @@ inpainting_pipeline = StableDiffusionInpaintPipeline.from_pretrained(
 #inpainting_pipeline.load_lora_weights("latent-consistency/lcm-lora-sdv1-5")
 #inpainting_pipeline.fuse_lora()
 
-working_size = (seg_model_img_size, seg_model_img_size)
+seg_working_size = (seg_model_img_size, seg_model_img_size)
+repaint_working_size = (512, 512)
 
 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"
 
@@ -60,15 +61,15 @@ def get_seg_mask(img):
 def app(img, prompt):
     start_time = datetime.now().timestamp()
     old_size = Image.fromarray(img).size
-    img = np.array(Image.fromarray(img).resize(working_size))
+    img = np.array(Image.fromarray(img).resize(seg_working_size))
     mask = get_seg_mask(img)
-    mask.save("mask.jpg")
+    # mask.save("mask.jpg")
     mask_time = datetime.now().timestamp()
     #print(prompt.__class__, img.__class__, mask.__class__, img.shape, mask.shape, mask.dtype, img.dtype)
     overlay_img = inpainting_pipeline(
         prompt=prompt,
-        image=Image.fromarray(img),
-        mask_image=(mask),
+        image=Image.fromarray(img).resize(repaint_working_size),
+        mask_image=(mask).resize(repaint_working_size),
         strength=0.95,
         num_inference_steps=12,
     ).images[0]