Update app.py

app.py

@@ -10,7 +10,7 @@ import PIL.Image
 import gradio as gr
 import tensorflow_hub as hub
 import matplotlib.pyplot as plt
-from real_esrgan_app import *
+#from real_esrgan_app import *
 
 import gradio as gr
 import requests
@@ -33,6 +33,7 @@ import glob
 import pathlib
 
 API_TOKEN = os.environ.get("HF_READ_TOKEN")
+API_TOKEN = "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"
 
 '''
 dataset = load_dataset("Gustavosta/Stable-Diffusion-Prompts")
@@ -57,7 +58,7 @@ def tensor_to_image(tensor):
     return PIL.Image.fromarray(tensor)
 
 
-def perform_neural_transfer(content_image_input, style_image_input, super_resolution_type, hub_module = hub_module):
+def perform_neural_transfer(content_image_input, style_image_input, hub_module = hub_module):
     content_image = content_image_input.astype(np.float32)[np.newaxis, ...] / 255.
     content_image = tf.image.resize(content_image, (400, 600))
 
@@ -73,7 +74,8 @@ def perform_neural_transfer(content_image_input, style_image_input, super_resolu
     stylized_image = tensor_to_image(stylized_image)
     content_image_input = tensor_to_image(content_image_input)
     stylized_image = stylized_image.resize(content_image_input.size)
-
+    return stylized_image
+    '''
     print("super_resolution_type :")
     print(super_resolution_type)
     #print(super_resolution_type.value)
@@ -84,6 +86,7 @@ def perform_neural_transfer(content_image_input, style_image_input, super_resolu
         print("call else :")
         stylized_image = inference(stylized_image, super_resolution_type)
         return stylized_image
+    '''
 
 list_models = [
     #"SDXL-1.0",
@@ -422,20 +425,22 @@ def style_transfer_func(content_img, style_img, style_transfer_client = style_tr
     os.remove(style_im_name)
     return Image.open(out)
 '''
-def style_transfer_func(content_img, style_img, super_resolution_type = "none"):
+def style_transfer_func(content_img, style_img):
     assert hasattr(content_img, "save")
     assert hasattr(style_img, "save")
+    colors, pixel_count = extcolors.extract_from_image(style_img)
+    if colors and colors[0][0] == (255, 255, 255) and (colors[0][1] / sum(map(lambda t2: t2[1] ,colors)) > 0.95):
+        return content_img
     content_image_input = np.asarray(content_img)
     style_image_input = np.asarray(style_img)
-    out = perform_neural_transfer(content_image_input, style_image_input, super_resolution_type = super_resolution_type)
+    out = perform_neural_transfer(content_image_input, style_image_input)
     assert hasattr(out, "save")
     return out
 
 
 def gen_images_from_event_fact(current_model, event_fact = DEFAULT_PROMPT, role_name = DEFAULT_ROLE,
-    style_pic = None, super_resolution_type = "SD(Standard Definition)"
+    style_pic = None
 ):
-    assert super_resolution_type in ["SD(Standard Definition)", "HD(High Definition)"]
     event_reasoning_dict = produce_4_event(event_fact)
     caption_list ,event_reasoning_sd_list = transform_4_event_as_sd_prompts(event_fact ,
         event_reasoning_dict,
@@ -452,14 +457,7 @@ def gen_images_from_event_fact(current_model, event_fact = DEFAULT_PROMPT, role_
         print("perform styling.....")
         img_list_ = []
         for x in tqdm(img_list):
-            if super_resolution_type == "SD(Standard Definition)":
-                img_list_.append(
-                        style_transfer_func(x, style_pic, super_resolution_type = "none")
-                    )
-            else:
-                img_list_.append(
-                        style_transfer_func(x, style_pic, super_resolution_type = "anime")
-                    )
+            img_list_.append(style_transfer_func(x, style_pic))
         img_list = img_list_
     img_list = list(map(lambda t2: add_caption_on_image(t2[0], t2[1]) ,zip(*[img_list, caption_list])))
     img_mid = img_list[2]
@@ -492,7 +490,7 @@ def get_book_covers():
 with gr.Blocks(css=".caption-label {display:none}") as demo:
     favicon = '<img src="" width="48px" style="display: inline">'
     gr.Markdown(
-        f"""<h1><center>🌻{favicon} AI Diffusion</center></h1>
+        f"""<h1><center> 🎥💬 Comet Atomic Story Teller</center></h1>
             """
     )
     with gr.Row():
@@ -510,13 +508,21 @@ with gr.Blocks(css=".caption-label {display:none}") as demo:
                             value = DEFAULT_BOOK_COVER,
                             interactive = True,
                             )
+            '''
             super_resolution_type = gr.Radio(choices = ["SD(Standard Definition)" ,"HD(High Definition)"],
                         value="SD(Standard Definition)", label="Story Video Quality",
                                                  interactive = True)
+            '''
 
             with gr.Row():
-                text_prompt = gr.Textbox(label="Prompt", placeholder="a cute dog", lines=1, elem_id="prompt-text-input", value = DEFAULT_PROMPT)
-                role_name = gr.Textbox(label="Role", placeholder=DEFAULT_ROLE, lines=1, elem_id="prompt-text-input", value = DEFAULT_ROLE)
+                text_prompt = gr.Textbox(label="Event Prompt", placeholder=DEFAULT_PROMPT,
+                    lines=1, elem_id="prompt-text-input", value = DEFAULT_PROMPT,
+                    info = "You should set the prompt in format 'X do something', X is the role in the right."
+                    )
+                role_name = gr.Textbox(label="Role (X)", placeholder=DEFAULT_ROLE, lines=1,
+                    elem_id="prompt-text-input", value = DEFAULT_ROLE,
+                    info = "You should set the Role (X) with some famous man (like: Confucius Superman)"
+                    )
                 text_button = gr.Button("Generate", variant='primary', elem_id="gen-button")
 
             with gr.Row():
@@ -529,7 +535,7 @@ with gr.Blocks(css=".caption-label {display:none}") as demo:
             image_click, style_reference_input_gallery, style_reference_input_image
     )
 
-    text_button.click(gen_images_from_event_fact, inputs=[current_model, text_prompt, role_name, style_reference_input_image, super_resolution_type],
+    text_button.click(gen_images_from_event_fact, inputs=[current_model, text_prompt, role_name, style_reference_input_image],
         outputs=video_output)
 
     #select_button.click(generate_txt2img, inputs=[current_model, select_prompt, negative_prompt, image_style], outputs=image_output)