Update app_merged.py

app_merged.py

@@ -71,6 +71,9 @@ import spaces
 from huggingface_hub import hf_hub_download
 
 
+MAX_IMAGE_WIDTH = 2048
+IMAGE_FORMAT = "JPEG"
+
 
 
 client = httpx.Client(timeout=httpx.Timeout(10.0))  # Set timeout to 10 seconds
@@ -456,7 +459,7 @@ def enable_efficient_attention():
 def clear_memory():
     if torch.cuda.is_available():
         torch.cuda.empty_cache()
-        torch.cuda.synchronize()
+        #torch.cuda.synchronize()
 
 # Enable efficient attention
 enable_efficient_attention()
@@ -1808,9 +1811,9 @@ with gr.Blocks() as app:
             #     outputs=[result_gallery, output_bg],
             #     run_on_click=True, examples_per_page=1024
             # )
-        augment_prompt.click(generate_description, inputs=[prompt, extracted_fg], outputs=[prompt])
+        augment_prompt.click(generate_description, inputs=[prompt, extracted_fg], outputs=[prompt]).then(clear_memory, inputs=[], outputs=[])
         ips = [extracted_fg, prompt, image_width, image_height, num_samples, seed, steps, a_prompt, n_prompt, cfg, highres_scale, highres_denoise, lowres_denoise, bg_source]
-        relight_button.click(fn=process_relight, inputs=ips, outputs=[result_gallery])
+        relight_button.click(fn=process_relight, inputs=ips, outputs=[result_gallery]).then(clear_memory, inputs=[], outputs=[])
         example_quick_prompts.click(lambda x, y: ', '.join(y.split(', ')[:2] + [x[0]]), inputs=[example_quick_prompts, prompt], outputs=prompt, show_progress=False, queue=False)
         example_quick_subjects.click(lambda x: x[0], inputs=example_quick_subjects, outputs=prompt, show_progress=False, queue=False)
 
@@ -1860,7 +1863,7 @@ with gr.Blocks() as app:
             inputs=[dummy_image_for_outputs, width_slider, height_slider, overlap_percentage, num_inference_steps,
                     resize_option, custom_resize_percentage, prompt_fill, alignment_dropdown,
                     overlap_left, overlap_right, overlap_top, overlap_bottom],
-            outputs=[fill_result])
+            outputs=[fill_result]).then(clear_memory, inputs=[], outputs=[])
         # ).then(
         #     fn=lambda: gr.update(visible=True),
         #     inputs=None,
@@ -1876,7 +1879,7 @@ with gr.Blocks() as app:
             inputs=[dummy_image_for_outputs, width_slider, height_slider, overlap_percentage, num_inference_steps,
                     resize_option, custom_resize_percentage, prompt_fill, alignment_dropdown,
                     overlap_left, overlap_right, overlap_top, overlap_bottom],
-            outputs=[fill_result])
+            outputs=[fill_result]).then(clear_memory, inputs=[], outputs=[])
 
 
         def convert_to_pil(image):
@@ -1943,14 +1946,14 @@ with gr.Blocks() as app:
         def send_img(img_result):
                 return img_result
         
-        transfer_btn.click(send_img, [output_image], [input_fg])
+        transfer_btn.click(send_img, [output_image], [input_fg]).then(clear_memory, inputs=[], outputs=[])
 
                 
         generate_btn.click(
             fn=generate_image,
             inputs=[prompt_input, structure_image, style_image, depth_strength, style_strength],
             outputs=[output_image]
-        )
+        ).then(clear_memory, inputs=[], outputs=[]).then(clear_memory, inputs=[], outputs=[])
 
 if __name__ == "__main__":
     app.queue(default_concurrency_limit=3)