Spaces:
Sleeping
Sleeping
Update image_gen.py
Browse files- image_gen.py +75 -50
image_gen.py
CHANGED
|
@@ -1,8 +1,6 @@
|
|
| 1 |
# -----------------------
|
| 2 |
# Image Generation
|
| 3 |
# -----------------------
|
| 4 |
-
|
| 5 |
-
|
| 6 |
import os
|
| 7 |
import re
|
| 8 |
import time
|
|
@@ -87,15 +85,20 @@ def standardize_and_validate_image(file_path):
|
|
| 87 |
|
| 88 |
def generate_image(prompt_text, style, model="hf"):
|
| 89 |
"""
|
| 90 |
-
Generate an image from a text prompt using
|
| 91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 92 |
Args:
|
| 93 |
-
prompt_text (str): The text prompt for image generation.
|
| 94 |
-
style (str or None): The style of the image (used for HF and Gemini models).
|
| 95 |
-
model (str): Which model to use
|
| 96 |
-
|
|
|
|
| 97 |
Returns:
|
| 98 |
-
tuple:
|
| 99 |
"""
|
| 100 |
try:
|
| 101 |
if model == "pollinations_turbo":
|
|
@@ -108,60 +111,104 @@ def generate_image(prompt_text, style, model="hf"):
|
|
| 108 |
print(f"Error from image generation API: {response.status_code}")
|
| 109 |
return None, None
|
| 110 |
image_bytes = response.content
|
| 111 |
-
|
| 112 |
elif model == "gemini":
|
| 113 |
# For Google's Gemini model
|
| 114 |
try:
|
| 115 |
-
|
| 116 |
-
# Get API key from environment variable
|
| 117 |
g_api_key = os.getenv("GEMINI")
|
| 118 |
if not g_api_key:
|
| 119 |
logger.error("GEMINI_API_KEY not found in environment variables")
|
| 120 |
print("Google Gemini API key is missing. Please set the GEMINI_API_KEY environment variable.")
|
| 121 |
return None, None
|
| 122 |
-
|
| 123 |
# Initialize Gemini client
|
| 124 |
client = genai.Client(api_key=g_api_key)
|
| 125 |
-
|
| 126 |
# Enhance prompt with style
|
| 127 |
enhanced_prompt = f"image of {prompt_text} in {style} style, high quality, detailed illustration"
|
| 128 |
-
|
| 129 |
# Generate content
|
| 130 |
response = client.models.generate_content(
|
| 131 |
model="models/gemini-2.0-flash-exp",
|
| 132 |
contents=enhanced_prompt,
|
| 133 |
config=types.GenerateContentConfig(response_modalities=['Text', 'Image'])
|
| 134 |
)
|
| 135 |
-
|
| 136 |
# Extract image from response
|
| 137 |
for part in response.candidates[0].content.parts:
|
| 138 |
if part.inline_data is not None:
|
| 139 |
image = Image.open(BytesIO(part.inline_data.data))
|
| 140 |
-
|
| 141 |
# Convert to base64 string
|
| 142 |
buffered = io.BytesIO()
|
| 143 |
image.save(buffered, format="JPEG")
|
| 144 |
img_str = base64.b64encode(buffered.getvalue()).decode()
|
| 145 |
-
|
| 146 |
return image, img_str
|
| 147 |
-
|
| 148 |
# If no image was found in the response
|
| 149 |
logger.error("No image was found in the Gemini API response")
|
| 150 |
print("Gemini API didn't return an image")
|
| 151 |
return None, None
|
| 152 |
-
|
| 153 |
except ImportError:
|
| 154 |
logger.error("Google Gemini libraries not installed")
|
| 155 |
-
|
| 156 |
return None, None
|
| 157 |
-
|
| 158 |
except Exception as e:
|
| 159 |
logger.error(f"Gemini API error: {str(e)}")
|
| 160 |
print(f"Error from Gemini image generation: {str(e)}")
|
| 161 |
return None, None
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 165 |
enhanced_prompt = f"{prompt_text} in {style} style, high quality, detailed illustration"
|
| 166 |
model_id = "black-forest-labs/FLUX.1-dev"
|
| 167 |
api_url = f"https://api-inference.huggingface.co/models/{model_id}"
|
|
@@ -173,43 +220,21 @@ def generate_image(prompt_text, style, model="hf"):
|
|
| 173 |
return None, None
|
| 174 |
image_bytes = response.content
|
| 175 |
|
| 176 |
-
# For HF
|
| 177 |
if model != "gemini":
|
| 178 |
image = Image.open(io.BytesIO(image_bytes))
|
| 179 |
buffered = io.BytesIO()
|
| 180 |
image.save(buffered, format="JPEG")
|
| 181 |
img_str = base64.b64encode(buffered.getvalue()).decode()
|
| 182 |
return image, img_str
|
| 183 |
-
|
| 184 |
except Exception as e:
|
| 185 |
print(f"Error generating image: {e}")
|
| 186 |
logger.error(f"Image generation error: {str(e)}")
|
| 187 |
-
|
| 188 |
# Return a placeholder image in case of failure
|
| 189 |
return Image.new('RGB', (1024, 1024), color=(200,200,200)), None
|
| 190 |
|
| 191 |
-
def generate_image_with_retry(prompt_text, style, model="hf", max_retries=3):
|
| 192 |
-
"""
|
| 193 |
-
Attempt to generate an image using generate_image, retrying up to max_retries if needed.
|
| 194 |
-
Args:
|
| 195 |
-
prompt_text (str): The text prompt for image generation.
|
| 196 |
-
style (str or None): The style of the image (ignored for Pollinations Turbo).
|
| 197 |
-
model (str): Which model to use ("hf" or "pollinations_turbo").
|
| 198 |
-
max_retries (int): Maximum number of retries.
|
| 199 |
-
Returns:
|
| 200 |
-
tuple: The generated image and its Base64 string.
|
| 201 |
-
"""
|
| 202 |
-
for attempt in range(max_retries):
|
| 203 |
-
try:
|
| 204 |
-
if attempt > 0:
|
| 205 |
-
time.sleep(2 ** attempt)
|
| 206 |
-
return generate_image(prompt_text, style, model=model)
|
| 207 |
-
except Exception as e:
|
| 208 |
-
logger.error(f"Attempt {attempt+1} failed: {e}")
|
| 209 |
-
if attempt == max_retries - 1:
|
| 210 |
-
raise
|
| 211 |
-
return None, None
|
| 212 |
-
|
| 213 |
# edit image function
|
| 214 |
def edit_section_image(image_url: str, gemini_prompt: str):
|
| 215 |
"""
|
|
|
|
| 1 |
# -----------------------
|
| 2 |
# Image Generation
|
| 3 |
# -----------------------
|
|
|
|
|
|
|
| 4 |
import os
|
| 5 |
import re
|
| 6 |
import time
|
|
|
|
| 85 |
|
| 86 |
def generate_image(prompt_text, style, model="hf"):
|
| 87 |
"""
|
| 88 |
+
Generate an image from a text prompt using one of the following:
|
| 89 |
+
- Hugging Face's FLUX.1-dev
|
| 90 |
+
- Pollinations Turbo
|
| 91 |
+
- Google's Gemini
|
| 92 |
+
- Pexels API (for a real photo instead of AI)
|
| 93 |
+
|
| 94 |
Args:
|
| 95 |
+
prompt_text (str): The text prompt for image generation or search.
|
| 96 |
+
style (str or None): The style of the image (used for HF and Gemini models, ignored for Pexels).
|
| 97 |
+
model (str): Which model to use
|
| 98 |
+
("hf", "pollinations_turbo", "gemini", or "pexels").
|
| 99 |
+
|
| 100 |
Returns:
|
| 101 |
+
tuple: (PIL.Image, base64_string) or (None, None) on error.
|
| 102 |
"""
|
| 103 |
try:
|
| 104 |
if model == "pollinations_turbo":
|
|
|
|
| 111 |
print(f"Error from image generation API: {response.status_code}")
|
| 112 |
return None, None
|
| 113 |
image_bytes = response.content
|
| 114 |
+
|
| 115 |
elif model == "gemini":
|
| 116 |
# For Google's Gemini model
|
| 117 |
try:
|
|
|
|
|
|
|
| 118 |
g_api_key = os.getenv("GEMINI")
|
| 119 |
if not g_api_key:
|
| 120 |
logger.error("GEMINI_API_KEY not found in environment variables")
|
| 121 |
print("Google Gemini API key is missing. Please set the GEMINI_API_KEY environment variable.")
|
| 122 |
return None, None
|
| 123 |
+
|
| 124 |
# Initialize Gemini client
|
| 125 |
client = genai.Client(api_key=g_api_key)
|
| 126 |
+
|
| 127 |
# Enhance prompt with style
|
| 128 |
enhanced_prompt = f"image of {prompt_text} in {style} style, high quality, detailed illustration"
|
| 129 |
+
|
| 130 |
# Generate content
|
| 131 |
response = client.models.generate_content(
|
| 132 |
model="models/gemini-2.0-flash-exp",
|
| 133 |
contents=enhanced_prompt,
|
| 134 |
config=types.GenerateContentConfig(response_modalities=['Text', 'Image'])
|
| 135 |
)
|
| 136 |
+
|
| 137 |
# Extract image from response
|
| 138 |
for part in response.candidates[0].content.parts:
|
| 139 |
if part.inline_data is not None:
|
| 140 |
image = Image.open(BytesIO(part.inline_data.data))
|
|
|
|
| 141 |
# Convert to base64 string
|
| 142 |
buffered = io.BytesIO()
|
| 143 |
image.save(buffered, format="JPEG")
|
| 144 |
img_str = base64.b64encode(buffered.getvalue()).decode()
|
|
|
|
| 145 |
return image, img_str
|
| 146 |
+
|
| 147 |
# If no image was found in the response
|
| 148 |
logger.error("No image was found in the Gemini API response")
|
| 149 |
print("Gemini API didn't return an image")
|
| 150 |
return None, None
|
| 151 |
+
|
| 152 |
except ImportError:
|
| 153 |
logger.error("Google Gemini libraries not installed")
|
| 154 |
+
print("Google Gemini libraries not installed. Install with 'pip install google-genai'")
|
| 155 |
return None, None
|
| 156 |
+
|
| 157 |
except Exception as e:
|
| 158 |
logger.error(f"Gemini API error: {str(e)}")
|
| 159 |
print(f"Error from Gemini image generation: {str(e)}")
|
| 160 |
return None, None
|
| 161 |
+
|
| 162 |
+
elif model == "pexels":
|
| 163 |
+
# ---------- NEW BRANCH FOR PEXELS -----------
|
| 164 |
+
pexels_api_key = os.getenv("PEXELS_API_KEY")
|
| 165 |
+
if not pexels_api_key:
|
| 166 |
+
logger.error("PEXELS_API_KEY not found in environment variables")
|
| 167 |
+
print("Pexels API key is missing. Please set the PEXELS_API_KEY environment variable.")
|
| 168 |
+
return None, None
|
| 169 |
+
|
| 170 |
+
# Call Pexels search endpoint
|
| 171 |
+
# e.g. GET https://api.pexels.com/v1/search?query={prompt_text}&per_page=1
|
| 172 |
+
|
| 173 |
+
search_url = "https://api.pexels.com/v1/search"
|
| 174 |
+
headers_pexels = {
|
| 175 |
+
"Authorization": pexels_api_key
|
| 176 |
+
}
|
| 177 |
+
params = {
|
| 178 |
+
"query": prompt_text,
|
| 179 |
+
"per_page": 1
|
| 180 |
+
}
|
| 181 |
+
response = requests.get(search_url, headers=headers_pexels, params=params)
|
| 182 |
+
if response.status_code != 200:
|
| 183 |
+
logger.error(f"Pexels API error: {response.status_code}, {response.text}")
|
| 184 |
+
print(f"Error from Pexels API: {response.status_code}")
|
| 185 |
+
return None, None
|
| 186 |
+
|
| 187 |
+
data = response.json()
|
| 188 |
+
photos = data.get("photos", [])
|
| 189 |
+
if not photos:
|
| 190 |
+
logger.error("No photos found for the given prompt on Pexels")
|
| 191 |
+
print("No photos found on Pexels for this prompt.")
|
| 192 |
+
return None, None
|
| 193 |
+
|
| 194 |
+
# Take the first photo
|
| 195 |
+
photo = photos[0]
|
| 196 |
+
# We can pick "src" => "original" or "large2x", etc.
|
| 197 |
+
image_url = photo["src"].get("large2x") or photo["src"].get("original")
|
| 198 |
+
if not image_url:
|
| 199 |
+
logger.error("No suitable image URL found in Pexels photo object")
|
| 200 |
+
return None, None
|
| 201 |
+
|
| 202 |
+
# Download the image
|
| 203 |
+
img_resp = requests.get(image_url)
|
| 204 |
+
if img_resp.status_code != 200:
|
| 205 |
+
logger.error(f"Failed to download Pexels image from {image_url}")
|
| 206 |
+
return None, None
|
| 207 |
+
|
| 208 |
+
image_bytes = img_resp.content
|
| 209 |
+
|
| 210 |
+
else:
|
| 211 |
+
# Default to Hugging Face model
|
| 212 |
enhanced_prompt = f"{prompt_text} in {style} style, high quality, detailed illustration"
|
| 213 |
model_id = "black-forest-labs/FLUX.1-dev"
|
| 214 |
api_url = f"https://api-inference.huggingface.co/models/{model_id}"
|
|
|
|
| 220 |
return None, None
|
| 221 |
image_bytes = response.content
|
| 222 |
|
| 223 |
+
# For HF, Pollinations, or Pexels that return image bytes
|
| 224 |
if model != "gemini":
|
| 225 |
image = Image.open(io.BytesIO(image_bytes))
|
| 226 |
buffered = io.BytesIO()
|
| 227 |
image.save(buffered, format="JPEG")
|
| 228 |
img_str = base64.b64encode(buffered.getvalue()).decode()
|
| 229 |
return image, img_str
|
| 230 |
+
|
| 231 |
except Exception as e:
|
| 232 |
print(f"Error generating image: {e}")
|
| 233 |
logger.error(f"Image generation error: {str(e)}")
|
| 234 |
+
|
| 235 |
# Return a placeholder image in case of failure
|
| 236 |
return Image.new('RGB', (1024, 1024), color=(200,200,200)), None
|
| 237 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 238 |
# edit image function
|
| 239 |
def edit_section_image(image_url: str, gemini_prompt: str):
|
| 240 |
"""
|