update inpaint

app.py

@@ -93,7 +93,16 @@ def text_to_image_generation(input_text, guidance_scale=1.75, generation_timeste
 
 
 @spaces.GPU
-def text_guided_inpainting(input_text, inpainting_image, inpainting_mask, guidance_scale=1.75, generation_timesteps=16):
+def text_guided_inpainting(input_text, inpainting_image, inpainting_mask_input, guidance_scale=1.75, generation_timesteps=16):
+    # pre-process inpainting mask
+    alpha_channel = inpainting_mask_input["layers"][0][:, :, 3]
+    mask = np.where(alpha_channel == 0, 0, 255).astype(np.uint8)
+
+    if np.sum(mask) == 0:
+        inpainting_mask = Image.fromarray(inpainting_mask_input['background']).convert('L')
+    else:
+        inpainting_mask = Image.fromarray(mask).convert('L')
+
     prompt = [input_text]
 
     config.training.batch_size = config.batch_size = 1
@@ -393,15 +402,23 @@ with gr.Blocks() as demo:
         inpainting_input_img = gr.Image(
             label="Input image",
             type="pil",
-            height=256,
-            width=256,
+            # height=256,
+            # width=256,
         )
-        inpainting_input_mask = gr.Image(
+        # inpainting_input_mask = gr.Image(
+        #     label="Inpainting mask",
+        #     image_mode="L",
+        #     type="pil",
+        #     height=256,
+        #     width=256,
+        # )
+        inpainting_input_mask = gr.ImageMask(
+            sources=["upload"],
+            layers=False,
+            transforms=[],
+            format="png",
             label="Inpainting mask",
-            image_mode="L",
-            type="pil",
-            height=256,
-            width=256,
+            show_label=True
         )
 
         with gr.Column():