feat: Agregar función de refinamiento de imágenes en app.py

Se ha agregado una nueva función llamada `refine_image` en el archivo app.py. Esta función permite aplicar un refinamiento a las imágenes generadas utilizando un modelo de refinería. Los parámetros de refinamiento, como la fuerza, el número de pasos de inferencia y la escala de guía, se pueden ajustar según las necesidades del usuario. Esta mejora proporciona mayor flexibilidad y control sobre el proceso de generación de imágenes.

app.py

@@ -5,6 +5,7 @@ import json
 from gradio_client import Client, handle_file
 from gradio_imageslider import ImageSlider
 from PIL import Image
+client = InferenceClient()
 
 with open('loras.json', 'r') as f:
     loras = json.load(f)
@@ -168,6 +169,21 @@ def upscale_image(image, resolution, num_inference_steps, strength, hdr, guidanc
     return result
 
 
+def refine_image(apply_refiner, image, model ,prompt, negative_prompt, num_inference_steps, guidance_scale, seed, strength):
+    if (not apply_refiner):
+        return [image, image]
+    
+    refined_image = client.image_to_image(
+        image, 
+        prompt=prompt, 
+        negative_prompt=negative_prompt, 
+        num_inference_steps=num_inference_steps, 
+        guidance_scale=guidance_scale, 
+        seed=seed, 
+        model=model,
+        strength=strength
+    )
+    return [image, refined_image]
 
 
 css="""
@@ -261,6 +277,50 @@ with gr.Blocks(css=css) as demo:
                         value=1.0
                     )
 
+                with gr.Group():
+                    apply_refiner = gr.Checkbox(label="Apply refiner", value=False)
+                    with gr.Accordion("Refiner params", open=False) as refiner_params:
+                        refiner_prompt = gr.Textbox(lines=3, label="Prompt")
+                        refiner_negative_prompt = gr.Textbox(lines=3, label="Negative Prompt")
+                        refiner_strength = gr.Slider(
+                            label="Strength",
+                            minimum=0,
+                            maximum=300,
+                            step=0.01,
+                            value=1
+                        )
+                        refiner_num_inference_steps = gr.Slider(
+                            label="Inference steps",
+                            minimum=3,
+                            maximum=300,
+                            step=1,
+                            value=25
+                        )
+                        refiner_guidance_scale = gr.Slider(
+                            label="Guidance scale",
+                            minimum=0.0,
+                            maximum=50.0,
+                            step=0.1,
+                            value=12
+                        )
+                        refiner_seed = gr.Slider(
+                            label="Seed",
+                            info="-1 denotes a random seed",
+                            minimum=-1,
+                            maximum=423538377342,
+                            step=1,
+                            value=-1
+                        )
+                        refiner_model = gr.Textbox(label="Model", value="stabilityai/stable-diffusion-xl-refiner-1.0")
+
+                    apply_refiner.change(
+                        fn=lambda x: gr.update(visible=x),
+                        inputs=apply_refiner,
+                        outputs=refiner_params,
+                        queue=False,
+                        api_name=False,
+                    )
+
         with gr.Column(scale=1):
             gallery = gr.Gallery(
                 [(item["image"], item["title"]) for item in loras],
@@ -282,7 +342,7 @@ with gr.Blocks(css=css) as demo:
         with gr.Column():
             output_slider = ImageSlider(label="Before / After", type="filepath", show_download_button=False)
 
-    with gr.Accordion("Advanced Options", open=False):
+    with gr.Accordion("Enhacer params", open=False):
         upscale_reduce_factor = gr.Slider(minimum=1, maximum=10, step=1, label="Reduce Factor", info="1/n")
         upscale_resolution = gr.Slider(minimum=128, maximum=2048, value=1024, step=128, label="Resolution", info="Image width")
         upscale_num_inference_steps = gr.Slider(minimum=1, maximum=150, value=50, step=1, label="Number of Inference Steps")
@@ -303,7 +363,11 @@ with gr.Blocks(css=css) as demo:
         fn=infer,
         inputs=[selected_index, prompt_in, style_prompt_in, inf_steps, guidance_scale, width, height, seed, lora_weight],
         outputs=[generated_image, last_used_seed, used_prompt]
+    ).then(refine_image, 
+        [apply_refiner, generated_image, refiner_model, refiner_prompt, refiner_negative_prompt, refiner_num_inference_steps, refiner_guidance_scale, refiner_seed, refiner_strength],
+        generated_image
     )
+
     cancel_btn.click(
         fn=cancel_infer,
         outputs=[]