Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -86,56 +86,115 @@ def augment_prompt_with_llm(prompt):
|
|
86 |
print(f"Error calling LLM API: {e}")
|
87 |
return prompt
|
88 |
|
89 |
-
def
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
# Create a copy of the image to work with
|
93 |
img_with_text = image.copy()
|
94 |
draw = ImageDraw.Draw(img_with_text)
|
95 |
|
96 |
-
#
|
97 |
-
|
98 |
-
|
99 |
-
title_font_size = 48
|
100 |
-
author_font_size = 32
|
101 |
-
# For production, you'd want to include proper font files
|
102 |
-
title_font = ImageFont.truetype("/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf", title_font_size)
|
103 |
-
author_font = ImageFont.truetype("/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf", author_font_size)
|
104 |
-
except:
|
105 |
-
# Fallback to default font
|
106 |
-
title_font = ImageFont.load_default()
|
107 |
-
author_font = ImageFont.load_default()
|
108 |
|
109 |
# Get image dimensions
|
110 |
img_width, img_height = img_with_text.size
|
111 |
|
112 |
# Define position mappings
|
113 |
position_coords = {
|
114 |
-
"
|
115 |
-
"
|
116 |
-
"
|
117 |
}
|
118 |
|
119 |
-
# Draw title
|
120 |
-
if title_ko
|
121 |
-
title_text = f"{title_ko}\n{title_en}" if title_ko and title_en else title_ko or title_en
|
122 |
title_x, title_y = position_coords[title_position]
|
123 |
|
124 |
# Get text bbox for centering
|
125 |
-
bbox = draw.textbbox((0, 0),
|
126 |
text_width = bbox[2] - bbox[0]
|
127 |
text_height = bbox[3] - bbox[1]
|
128 |
|
129 |
# Draw text with shadow for better visibility
|
130 |
shadow_offset = 2
|
131 |
draw.text((title_x - text_width // 2 + shadow_offset, title_y - text_height // 2 + shadow_offset),
|
132 |
-
|
133 |
draw.text((title_x - text_width // 2, title_y - text_height // 2),
|
134 |
-
|
135 |
|
136 |
-
# Draw author
|
137 |
-
if author_ko
|
138 |
-
|
|
|
|
|
|
|
|
|
|
|
139 |
author_x, author_y = position_coords[author_position]
|
140 |
|
141 |
# Get text bbox for centering
|
@@ -145,9 +204,9 @@ def add_text_overlay(image, title_ko, title_en, author_ko, author_en,
|
|
145 |
|
146 |
# Draw text with shadow
|
147 |
draw.text((author_x - text_width // 2 + shadow_offset, author_y - text_height // 2 + shadow_offset),
|
148 |
-
author_text, font=author_font, fill="black"
|
149 |
draw.text((author_x - text_width // 2, author_y - text_height // 2),
|
150 |
-
author_text, font=author_font, fill=text_color
|
151 |
|
152 |
return img_with_text
|
153 |
|
@@ -194,12 +253,13 @@ def inference(
|
|
194 |
num_inference_steps: int,
|
195 |
lora_scale: float,
|
196 |
title_ko: str,
|
197 |
-
title_en: str,
|
198 |
author_ko: str,
|
199 |
-
author_en: str,
|
200 |
title_position: str,
|
201 |
author_position: str,
|
202 |
text_color: str,
|
|
|
|
|
|
|
203 |
progress: gr.Progress = gr.Progress(track_tqdm=True),
|
204 |
):
|
205 |
if randomize_seed:
|
@@ -216,10 +276,11 @@ def inference(
|
|
216 |
joint_attention_kwargs={"scale": lora_scale},
|
217 |
).images[0]
|
218 |
|
219 |
-
# Add text overlay if any text is provided
|
220 |
-
if
|
221 |
-
image = add_text_overlay(image, title_ko,
|
222 |
-
title_position, author_position, text_color
|
|
|
223 |
|
224 |
# Save the generated image
|
225 |
filepath = save_generated_image(image, prompt)
|
@@ -278,29 +339,47 @@ with gr.Blocks(theme=gr.themes.Soft(), analytics_enabled=False) as demo:
|
|
278 |
value=DEFAULT_IMAGE_PATH # Set the default image
|
279 |
)
|
280 |
|
281 |
-
with gr.Accordion("Text Overlay Settings", open=False):
|
282 |
with gr.Row():
|
283 |
with gr.Column():
|
284 |
-
title_ko = gr.Textbox(label="
|
285 |
-
title_en = gr.Textbox(label="Title (English)", placeholder="English Title")
|
286 |
title_position = gr.Radio(
|
287 |
-
label="
|
288 |
-
choices=["
|
289 |
-
value="
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
290 |
)
|
291 |
with gr.Column():
|
292 |
-
author_ko = gr.Textbox(label="
|
293 |
-
author_en = gr.Textbox(label="Author (English)", placeholder="Author Name")
|
294 |
author_position = gr.Radio(
|
295 |
-
label="
|
296 |
-
choices=["
|
297 |
-
value="
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
298 |
)
|
299 |
|
300 |
-
|
301 |
-
|
302 |
-
|
303 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
304 |
|
305 |
with gr.Accordion("Advanced Settings", open=False):
|
306 |
seed = gr.Slider(
|
@@ -412,12 +491,13 @@ with gr.Blocks(theme=gr.themes.Soft(), analytics_enabled=False) as demo:
|
|
412 |
num_inference_steps,
|
413 |
lora_scale,
|
414 |
title_ko,
|
415 |
-
title_en,
|
416 |
author_ko,
|
417 |
-
author_en,
|
418 |
title_position,
|
419 |
author_position,
|
420 |
text_color,
|
|
|
|
|
|
|
421 |
],
|
422 |
outputs=[result, seed, generated_gallery],
|
423 |
)
|
|
|
86 |
print(f"Error calling LLM API: {e}")
|
87 |
return prompt
|
88 |
|
89 |
+
def get_korean_font(font_name, font_size):
|
90 |
+
"""Get Korean font based on font name"""
|
91 |
+
font_paths = {
|
92 |
+
"λλκ³ λ": [
|
93 |
+
"/usr/share/fonts/truetype/nanum/NanumGothic.ttf",
|
94 |
+
"NanumGothic.ttf",
|
95 |
+
"/System/Library/Fonts/AppleSDGothicNeo.ttc", # macOS
|
96 |
+
"C:/Windows/Fonts/NanumGothic.ttf", # Windows
|
97 |
+
],
|
98 |
+
"λλλͺ
μ‘°": [
|
99 |
+
"/usr/share/fonts/truetype/nanum/NanumMyeongjo.ttf",
|
100 |
+
"NanumMyeongjo.ttf",
|
101 |
+
"C:/Windows/Fonts/NanumMyeongjo.ttf", # Windows
|
102 |
+
],
|
103 |
+
"λ§μ κ³ λ": [
|
104 |
+
"/usr/share/fonts/truetype/malgun/malgun.ttf",
|
105 |
+
"malgun.ttf",
|
106 |
+
"/System/Library/Fonts/AppleMyungjo.ttf", # macOS
|
107 |
+
"C:/Windows/Fonts/malgun.ttf", # Windows
|
108 |
+
],
|
109 |
+
"λ°ν": [
|
110 |
+
"/usr/share/fonts/truetype/batang/batang.ttf",
|
111 |
+
"batang.ttf",
|
112 |
+
"C:/Windows/Fonts/batang.ttc", # Windows
|
113 |
+
],
|
114 |
+
"λμ": [
|
115 |
+
"/usr/share/fonts/truetype/dotum/dotum.ttf",
|
116 |
+
"dotum.ttf",
|
117 |
+
"C:/Windows/Fonts/dotum.ttc", # Windows
|
118 |
+
],
|
119 |
+
"κΈ°λ³Έ": []
|
120 |
+
}
|
121 |
+
|
122 |
+
# Try to load the selected font
|
123 |
+
if font_name in font_paths:
|
124 |
+
for font_path in font_paths[font_name]:
|
125 |
+
try:
|
126 |
+
return ImageFont.truetype(font_path, font_size)
|
127 |
+
except:
|
128 |
+
continue
|
129 |
+
|
130 |
+
# Try common font directories
|
131 |
+
common_paths = [
|
132 |
+
f"/usr/share/fonts/truetype/{font_name.lower()}/{font_name}.ttf",
|
133 |
+
f"fonts/{font_name}.ttf",
|
134 |
+
f"{font_name}.ttf",
|
135 |
+
]
|
136 |
+
|
137 |
+
for path in common_paths:
|
138 |
+
try:
|
139 |
+
return ImageFont.truetype(path, font_size)
|
140 |
+
except:
|
141 |
+
continue
|
142 |
+
|
143 |
+
# Fallback to default font with larger size
|
144 |
+
try:
|
145 |
+
return ImageFont.truetype("DejaVuSans.ttf", font_size)
|
146 |
+
except:
|
147 |
+
# Use default font but try to make it larger
|
148 |
+
default = ImageFont.load_default()
|
149 |
+
# PIL default font is very small, so we might need to scale the text
|
150 |
+
return default
|
151 |
+
|
152 |
+
def add_text_overlay(image, title_ko, author_ko,
|
153 |
+
title_position, author_position, text_color, font_name,
|
154 |
+
title_size, author_size):
|
155 |
+
"""Add Korean text overlay to the generated image"""
|
156 |
# Create a copy of the image to work with
|
157 |
img_with_text = image.copy()
|
158 |
draw = ImageDraw.Draw(img_with_text)
|
159 |
|
160 |
+
# Load Korean fonts with custom sizes
|
161 |
+
title_font = get_korean_font(font_name, title_size)
|
162 |
+
author_font = get_korean_font(font_name, author_size)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
163 |
|
164 |
# Get image dimensions
|
165 |
img_width, img_height = img_with_text.size
|
166 |
|
167 |
# Define position mappings
|
168 |
position_coords = {
|
169 |
+
"μλ¨": (img_width // 2, img_height // 10),
|
170 |
+
"μ€μ": (img_width // 2, img_height // 2),
|
171 |
+
"νλ¨": (img_width // 2, img_height * 9 // 10)
|
172 |
}
|
173 |
|
174 |
+
# Draw title (Korean only)
|
175 |
+
if title_ko:
|
|
|
176 |
title_x, title_y = position_coords[title_position]
|
177 |
|
178 |
# Get text bbox for centering
|
179 |
+
bbox = draw.textbbox((0, 0), title_ko, font=title_font)
|
180 |
text_width = bbox[2] - bbox[0]
|
181 |
text_height = bbox[3] - bbox[1]
|
182 |
|
183 |
# Draw text with shadow for better visibility
|
184 |
shadow_offset = 2
|
185 |
draw.text((title_x - text_width // 2 + shadow_offset, title_y - text_height // 2 + shadow_offset),
|
186 |
+
title_ko, font=title_font, fill="black")
|
187 |
draw.text((title_x - text_width // 2, title_y - text_height // 2),
|
188 |
+
title_ko, font=title_font, fill=text_color)
|
189 |
|
190 |
+
# Draw author (Korean only)
|
191 |
+
if author_ko:
|
192 |
+
# Add "μ§μμ΄: " prefix if not already present
|
193 |
+
if not author_ko.startswith("μ§μμ΄"):
|
194 |
+
author_text = f"μ§μμ΄: {author_ko}"
|
195 |
+
else:
|
196 |
+
author_text = author_ko
|
197 |
+
|
198 |
author_x, author_y = position_coords[author_position]
|
199 |
|
200 |
# Get text bbox for centering
|
|
|
204 |
|
205 |
# Draw text with shadow
|
206 |
draw.text((author_x - text_width // 2 + shadow_offset, author_y - text_height // 2 + shadow_offset),
|
207 |
+
author_text, font=author_font, fill="black")
|
208 |
draw.text((author_x - text_width // 2, author_y - text_height // 2),
|
209 |
+
author_text, font=author_font, fill=text_color)
|
210 |
|
211 |
return img_with_text
|
212 |
|
|
|
253 |
num_inference_steps: int,
|
254 |
lora_scale: float,
|
255 |
title_ko: str,
|
|
|
256 |
author_ko: str,
|
|
|
257 |
title_position: str,
|
258 |
author_position: str,
|
259 |
text_color: str,
|
260 |
+
font_name: str,
|
261 |
+
title_size: int,
|
262 |
+
author_size: int,
|
263 |
progress: gr.Progress = gr.Progress(track_tqdm=True),
|
264 |
):
|
265 |
if randomize_seed:
|
|
|
276 |
joint_attention_kwargs={"scale": lora_scale},
|
277 |
).images[0]
|
278 |
|
279 |
+
# Add text overlay if any Korean text is provided
|
280 |
+
if title_ko or author_ko:
|
281 |
+
image = add_text_overlay(image, title_ko, author_ko,
|
282 |
+
title_position, author_position, text_color, font_name,
|
283 |
+
title_size, author_size)
|
284 |
|
285 |
# Save the generated image
|
286 |
filepath = save_generated_image(image, prompt)
|
|
|
339 |
value=DEFAULT_IMAGE_PATH # Set the default image
|
340 |
)
|
341 |
|
342 |
+
with gr.Accordion("Text Overlay Settings (νκΈ)", open=False):
|
343 |
with gr.Row():
|
344 |
with gr.Column():
|
345 |
+
title_ko = gr.Textbox(label="μ λͺ©", placeholder="νκΈ μ λͺ©μ μ
λ ₯νμΈμ")
|
|
|
346 |
title_position = gr.Radio(
|
347 |
+
label="μ λͺ© μμΉ",
|
348 |
+
choices=["μλ¨", "μ€μ", "νλ¨"],
|
349 |
+
value="μλ¨"
|
350 |
+
)
|
351 |
+
title_size = gr.Slider(
|
352 |
+
label="μ λͺ© κΈμ ν¬κΈ°",
|
353 |
+
minimum=20,
|
354 |
+
maximum=100,
|
355 |
+
value=48,
|
356 |
+
step=2
|
357 |
)
|
358 |
with gr.Column():
|
359 |
+
author_ko = gr.Textbox(label="μ§μμ΄", placeholder="μ§μμ΄ μ΄λ¦μ μ
λ ₯νμΈμ")
|
|
|
360 |
author_position = gr.Radio(
|
361 |
+
label="μ§μμ΄ μμΉ",
|
362 |
+
choices=["μλ¨", "μ€μ", "νλ¨"],
|
363 |
+
value="νλ¨"
|
364 |
+
)
|
365 |
+
author_size = gr.Slider(
|
366 |
+
label="μ§μμ΄ κΈμ ν¬κΈ°",
|
367 |
+
minimum=16,
|
368 |
+
maximum=60,
|
369 |
+
value=32,
|
370 |
+
step=2
|
371 |
)
|
372 |
|
373 |
+
with gr.Row():
|
374 |
+
font_name = gr.Dropdown(
|
375 |
+
label="ν°νΈ μ ν",
|
376 |
+
choices=["λλκ³ λ", "λλλͺ
μ‘°", "λ§μ κ³ λ", "λ°ν", "λμ", "κΈ°λ³Έ"],
|
377 |
+
value="λλκ³ λ"
|
378 |
+
)
|
379 |
+
text_color = gr.ColorPicker(
|
380 |
+
label="κΈμ μμ",
|
381 |
+
value="#FFFFFF"
|
382 |
+
)
|
383 |
|
384 |
with gr.Accordion("Advanced Settings", open=False):
|
385 |
seed = gr.Slider(
|
|
|
491 |
num_inference_steps,
|
492 |
lora_scale,
|
493 |
title_ko,
|
|
|
494 |
author_ko,
|
|
|
495 |
title_position,
|
496 |
author_position,
|
497 |
text_color,
|
498 |
+
font_name,
|
499 |
+
title_size,
|
500 |
+
author_size,
|
501 |
],
|
502 |
outputs=[result, seed, generated_gallery],
|
503 |
)
|