update

app.py

@@ -78,17 +78,36 @@ def show_warning(warning_text: str) -> gr.Blocks:
             gr.Markdown(warning_text)
     return demo
 
+def set_example_image(example: list):
+    return gr.update(value=example[0])
 
 def create_inference_demo(func: inference_fn) -> gr.Blocks:
     with gr.Blocks() as demo:
         with gr.Row():
             with gr.Column():
+                exemplar_img = gr.Image(
+                    label='Exemplar Image',
+                    type='pil',
+                    interaction=False
+                )
+                # paths = sorted(pathlib.Path('exemplars').glob('*.jpg'))
+                # exemplar_dataset = gr.Dataset(components=[exemplar_img],
+                #                             samples=[[path.as_posix()]
+                #                                      for path in paths])
+                exemplar_dataset = gr.Dataset(
+                    components=[exemplar_img],
+                    samples = [
+                        ['exemplars/painted_on.jpg'],
+                        ['exemplars/carved_by.jpg'],
+                        ['exemplars/inside.jpg']
+                    ]
+                )
+
                 model_id = gr.Dropdown(
                     choices=['painted_on', 'carved_by', 'inside'],
                     value='painted_on',
                     label='Relation',
                     visible=True)
-                # reload_button = gr.Button('Reload Weight List')
                 prompt = gr.Textbox(
                     label='Prompt',
                     max_lines=1,
@@ -120,27 +139,49 @@ def create_inference_demo(func: inference_fn) -> gr.Blocks:
                 result = gr.Image(label='Result')
 
 
-        prompt.submit(fn=func,
-                      inputs=[
-                          model_id,
-                          prompt,
-                          num_samples,
-                          guidance_scale,
-                          ddim_steps
-                      ],
-                      outputs=result,
-                      queue=False)
-
-        run_button.click(fn=func,
-                        inputs=[
-                            model_id,
-                            prompt,
-                            num_samples,
-                            guidance_scale,
-                            ddim_steps
-                        ],
-                         outputs=result,
-                         queue=False)
+        exemplar_dataset.click(fn=set_example_image,
+                                inputs=exemplar_dataset,
+                                outputs=exemplar_dataset.components,
+                                queue=False)
+        prompt.submit(
+            fn=func,
+            # inputs=[
+            #     model_id,
+            #     prompt,
+            #     num_samples,
+            #     guidance_scale,
+            #     ddim_steps
+            # ],
+            inputs=[
+                exemplar_dataset,
+                prompt,
+                num_samples,
+                guidance_scale,
+                ddim_steps
+            ],
+            outputs=result,
+            queue=False
+        )
+
+        run_button.click(
+            fn=func,
+            # inputs=[
+            #     model_id,
+            #     prompt,
+            #     num_samples,
+            #     guidance_scale,
+            #     ddim_steps
+            # ],
+            inputs=[
+                exemplar_dataset,
+                prompt,
+                num_samples,
+                guidance_scale,
+                ddim_steps
+            ],
+            outputs=result,
+            queue=False
+        )
     return demo
 
 

inference.py

@@ -42,12 +42,14 @@ def make_image_grid(imgs, rows, cols):
 
 
 def inference_fn(
-        model_id: str,
+        examples: list,
         prompt: str,
         num_samples: int,
         guidance_scale: float,
         ddim_steps: int,
     ) -> PIL.Image.Image:
+    # select model_id
+    model_id = pathlib.Path(examples[0]).stem
 
     # create inference pipeline
     if torch.cuda.is_available():