Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+images/sample/output_sample/dahi_durga[[:space:]]puja_58861.png filter=lfs diff=lfs merge=lfs -text
+images/sample/output_sample/doi[[:space:]]dahi_durga[[:space:]]puja_92098.png filter=lfs diff=lfs merge=lfs -text
+images/sample/output_sample/doi[[:space:]]dahi_durga[[:space:]]puja_93931.png filter=lfs diff=lfs merge=lfs -text
+images/sample/output_sample/pepsi[[:space:]]and[[:space:]]coca[[:space:]]cola_durga[[:space:]]puja_red,[[:space:]]white,[[:space:]]green,[[:space:]]orange[[:space:]]and[[:space:]]blue_20[[:space:]]percent[[:space:]]discount_7819.png filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,18 @@
+# Use an official Python runtime as a base image
+FROM python:3.11.9
+
+# Set the working directory in the container
+WORKDIR /app
+
+# Copy requirements.txt and install dependencies
+COPY requirements.txt /app/
+RUN pip install -r requirements.txt
+
+# Copy the rest of the application
+COPY . /app/
+
+# Expose the ports where the apps will run
+EXPOSE 8000 7860
+
+# Start both backend and frontend
+CMD bash -c "uvicorn backend:app --host 0.0.0.0 --port 8000 && python frontend.py"

README.md

@@ -1,12 +1,8 @@
 ---
-title: Bigbasket Promo Generator
-emoji: 🚀
-colorFrom: purple
-colorTo: pink
+title: Bigbasket_Promo_Generator
+app_file: frontend.py
 sdk: gradio
-sdk_version: 4.44.1
-app_file: app.py
-pinned: false
+sdk_version: 4.44.0
 ---
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# bigbasket_promo_generator
+AI Powered Banner and Promo Generator for Bigbasket

backend.py

@@ -0,0 +1,55 @@
+from fastapi import FastAPI, UploadFile, Form
+from typing import List, Optional
+from banner_generator import BannerGenerator, info  # Import BannerGenerator and info
+import shutil
+import os
+from fastapi.responses import FileResponse
+
+app = FastAPI()
+
+# Ensure directories for saving images exist
+os.makedirs('images/temp', exist_ok=True)
+os.makedirs('images/output', exist_ok=True)
+
+# Endpoint for generating a banner
+@app.post("/generate_banner")
+async def generate_banner(
+    topic: str = Form(...),
+    images: List[UploadFile] = Form(...),
+    aspect_ratio: Optional[str] = Form(None)  # Added aspect_ratio as input parameter
+):
+    # Save uploaded images to disk
+    image_paths = []
+    for image in images:
+        temp_path = f"images/temp/{os.path.basename(image.filename)}"
+        try:
+            with open(temp_path, "wb") as buffer:
+                shutil.copyfileobj(image.file, buffer)
+            image_paths.append(temp_path)
+        except FileNotFoundError as e:
+            return {"error": f"Could not save image: {e}"}
+
+    # Create a BannerGenerator instance
+    banner_generator = BannerGenerator(
+        CONFIGS=info,
+        topic=topic,
+        images=image_paths,
+        aspect_ratio=aspect_ratio  # Pass the aspect_ratio parameter
+    )
+
+    # Execute the banner generation process
+    try:
+        final_image_path = banner_generator.execute()
+    except Exception as e:
+        return {"error": f"Banner generation failed: {e}"}
+
+    # Return the relative path to the generated banner image
+    return {"image_path": f"/images/{os.path.basename(final_image_path)}"}
+
+# Endpoint to serve images
+@app.get("/images/{image_name}")
+async def get_image(image_name: str):
+    image_path = os.path.join("images/output", image_name)
+    if not os.path.exists(image_path):
+        return {"error": "Image not found"}
+    return FileResponse(image_path)

banner_generator.py

@@ -0,0 +1,208 @@
+import os
+import configparser
+import PIL
+from PIL import Image
+from typing import Any, List, Optional, Literal
+from pydantic import BaseModel, Field
+import google.generativeai as genai
+import vertexai
+from vertexai.preview.vision_models import ImageGenerationModel, GeneratedImage
+from vertexai.preview.vision_models import Image as VertexImage
+
+# Load configuration
+config = configparser.ConfigParser()
+config.read('config.ini')
+
+api_key = config['credentials']['gemini_api_key']
+proj_id = config['credentials']['gcp_project_id']
+info = {"PROJECT_ID": proj_id, "LOCATION": "us-central1", "API_KEY": api_key}
+
+# Create directories for temporary and output images if they do not exist
+os.makedirs('images/temp', exist_ok=True)
+os.makedirs('images/output', exist_ok=True)
+
+class BannerGenerator(BaseModel):
+    """Class responsible for generating banners using generative models."""
+
+    CONFIGS: dict
+    topic: str
+    images: List[str] | None = Field(default=None)
+    aspect_ratio: Optional[Literal["1:1", "9:16", "16:9", "4:3", "3:4"]] = None  # Added aspect_ratio attribute
+    text_model: str = "gemini-1.5-flash"
+    image_model: str = "imagen-3.0-generate-001"
+    edit_model: str = "imagegeneration@006"
+    pm: Any = None
+    im: Any = None
+    em: Any = None
+    text_v0: str = None
+    text_v1: str = None
+    text_v2: str = None
+    text_v3: str = None
+    img_response_v1: Any = None
+    img_response_v2: Any = None
+    img_response_v3: Any = None
+    launch_state: bool = False
+
+    def __launch(self):
+        """Launches the generative models and sets up environment."""
+        if not self.launch_state:
+            vertexai.init(project=self.CONFIGS['PROJECT_ID'], location=self.CONFIGS['LOCATION'])
+            genai.configure(api_key=self.CONFIGS['API_KEY'])
+            self.pm = genai.GenerativeModel(self.text_model)
+            self.im = ImageGenerationModel.from_pretrained(self.image_model)
+            self.em = ImageGenerationModel.from_pretrained(self.edit_model)
+            self.launch_state = True
+            print("Model Launch successful!")
+
+    def load_images(self) -> List[PIL.Image.Image]:
+        """Loads images from file paths provided in the `images` attribute."""
+        self.__launch()
+        loaded_images = []
+
+        for image_path in self.images:
+            img = PIL.Image.open(image_path)
+            if img.mode == 'RGBA':
+                img = img.convert('RGB')  # Convert to RGB if needed
+            loaded_images.append(img)
+
+        return loaded_images
+
+    def extract_image_information(self) -> str:
+        """Extracts information from images using the generative text model."""
+        images = self.load_images()
+        extraction_prompt = '''Examine the set of images to extract information about the product (name, logo) in less than 80 words.'''
+        model_input = [extraction_prompt] + images
+        response = self.pm.generate_content(model_input)
+        print("Attached images examined!")
+        return response.text
+
+    def extract_information(self) -> None:
+        """Extracts information from the given topic and images using a detailed analysis."""
+        self.__launch()
+
+        out_text = f"""Deep analyze text from retail advertising, marketing psychology, and thoroughly researched marketing studies perspective: {self.topic}
+        Extract the following information:
+        0. Product: Product name, brand and supplier, logo, tagline, size, packaging if available
+        1. Objective: 1 word for primary goal of the banner. Example - Awareness, Engagement, Conversion, Branding
+        2. Festival: Event or occasion it may be tied to. Example - Christmas, Diwali, Black Friday, Summer Sale, New Year, Generic
+        3. Headline: Suggest a main text that captures attention. Example - Discover [product] for [festival], Shop now and save!, Limited time offer, Innovate your life with [product]
+        4. Subheadline: Optional additional supporting information to clarify the offer. Example - Get 50% off until [date], Exclusive deal for festive season, Hurry offer ends soon
+        5. CTA: Add a call to action. Example - Buy now, Shop the collection, Discover More, Sign up today
+        6. Color Scheme: Use color palette based on audience, occasion, or product tone. Example - Red & Gold (Festive, Urgency), Blue & White (Trust, Calm), Green & Brown (Eco-friendly, Natural), Black & White (Elegant, Minimal)
+        7. Promotional offer: Suggest 1 best promotional offer. Example - MAX ₹99 OFF, UP TO 60% OFF, UNDER ₹999, MIN ₹10 OFF, MIN 20% OFF, STARTS @₹99, FLAT ₹100 OFF, FLAT 20% OFF, ₹499 STORE, BUY 2 GET 1 FREE
+        8. Background color gradient: Dynamic color generation to match overall look and feel
+        9. Background theme: Festival oriented or generic if no festival
+        """
+        self.text_v0 = self.pm.generate_content(out_text).text
+
+        # Information consolidation
+        out_text = f"Respond concisely and summarize in Python dictionary format only this: {self.text_v0}"
+        if self.images:
+            image_info = self.extract_image_information()
+            out_text += ' Product insights: ' + image_info
+            print(f"Product insights: {image_info}")
+
+        self.text_v1 = self.pm.generate_content(out_text).text[9:-5]
+
+        # Scrapper to ensure data integrity and consistency
+        out_text = f"Respond concisely by scrapping all unavailable information in Python dictionary format only this: {self.text_v1}"
+        self.text_v2 = self.pm.generate_content(out_text).text
+
+        print("Information collection complete!")
+
+    def create_text_prompt(self) -> None:
+        """Creates a text prompt based on the extracted information."""
+        out_text = f"""Task: Fill in the values in this json: {self.text_v2}
+        Guidelines:
+        1. It will be used to generate an ads banner.
+        2. Ensure it has all details pair-wise meticulously captured.
+        3. All unknown/missing/unprovided variables are replaced with the attributes of the most probable shopper for that product.
+        4. Recheck and identify all ambiguity or any text that leads to uncertainty.
+        5. Replace all uncertainty with targeted values that make the most sense for the given product.
+        6. Quantify everything possible, like high, medium, and lows to percentage values based on marketing and psychometric research studies.
+        7. All KPIs and qualitative measures are to be used subcontextually only. Remove any details about statistical testing or names of any performance KPIs.
+        8. Avoid sentences and use only necessary keywords.
+        9. Remove all redundant key-value pairs.
+        """
+        self.text_v3 = self.pm.generate_content(out_text).text
+        print("Information processed!")
+
+    def generate_image(self) -> str:
+        """Generates an image based on the given prompt and saves it in images/temp/."""
+        prompt = f"""Realistic, subcontextually implied qualitative attributes inspired, excellent image quality ad capturing every detail in json:{self.text_v3}"""
+
+        # Adding the aspect ratio parameter to the image generation
+        if self.aspect_ratio:
+            self.img_response_v1 = self.im.generate_images(prompt=prompt, aspect_ratio=self.aspect_ratio)
+        else:
+            self.img_response_v1 = self.im.generate_images(prompt=prompt)
+
+        # Save the generated image to images/temp/
+        temp_img_path = 'images/temp/temp.jpg'
+        if os.path.exists(temp_img_path):
+            os.remove(temp_img_path)
+        self.img_response_v1.images[0].save(temp_img_path)
+
+        print("Image v1 generated!")
+        return temp_img_path
+
+    def identify_lags(self) -> str:
+        """Identifies quality issues in the generated image and provides suggestions for improvement."""
+        prompt = f"""Be direct. Quality check the banner out of 10 on:
+        1. Promotional offer present as per instructions below
+        2. Ensure ALL texts pass grammatical checks
+        3. Color pallette as per instructions below
+        4. Occassion or festival theme is present as per instructions below
+
+        ONLY USE INFORMATION FROM {self.text_v3}. Don't DO NOT make up colors, promo or occasion. Make sure the promo and color pallete is followed as per above instructions.
+
+        Precisely point out errors and corresponding actions to fix the image where score is below 8.
+        Do not output anything about elements that need no change.
+        """
+        temp_img_path = 'images/temp/temp.jpg'
+        response = self.pm.generate_content([prompt, PIL.Image.open(temp_img_path)])
+        print(f'Lags identified: {response.text}')
+        return response.text
+
+    def fix_image(self, retest: bool = False) -> str:
+        """Attempts to fix the identified lags in the generated image and saves it."""
+        prompt = f'Realistic, subcontextually implied qualitative attributes inspired, excellent image quality ad by: {self.identify_lags()}'
+        temp_img_path = 'images/temp/temp.jpg'
+
+        base_image = VertexImage.load_from_file(location=temp_img_path)
+        self.img_response_v2 = self.em.edit_image(
+            base_image=base_image,
+            prompt=prompt,
+            edit_mode="inpainting-insert",
+            mask_mode="background"
+        )
+        self.img_response_v2.images[0].save(temp_img_path)
+        print("Image v2 generated!")
+
+        if retest:
+            prompt = f'Realistic, subcontextually implied qualitative attributes inspired, excellent image quality ad edit by: {self.identify_lags()}'
+            self.img_response_v3 = self.em.edit_image(
+                base_image=VertexImage.load_from_file(location=temp_img_path),
+                prompt=prompt,
+                edit_mode="inpainting-insert",
+                mask_mode="background"
+            )
+            self.img_response_v3.images[0].save(temp_img_path)
+            print("Image v3 generated!")
+
+        # Save final image to output directory with a name corresponding to the topic
+        output_filename = ''.join(e for e in self.topic if e.isalnum() or e == ' ')
+        output_filename = output_filename[:50].strip().replace(' ', '_') + ".jpg"
+        output_img_path = os.path.join('images/output', output_filename)
+        self.img_response_v2.images[0].save(output_img_path)
+        print(f"Final image saved as: {output_img_path}")
+
+        return output_img_path
+
+    def execute(self, QC=False) -> str:
+        """Executes the entire workflow to generate and refine the banner."""
+        self.extract_information()
+        self.create_text_prompt()
+        self.generate_image()
+        final_image_path = self.fix_image(retest=QC)
+        return final_image_path

frontend.py

@@ -0,0 +1,129 @@
+import gradio as gr
+import requests
+import os
+
+# BASE_URL = "http://localhost:8000"
+BASE_URL = os.getenv("BASE_URL", "http://localhost:8000")  # Default to localhost for local testing
+
+# Function to generate the banner
+def generate_banner(topic, images, aspect_ratio):
+    # Preparing files for the request
+    files = [('images', (img.name, open(img.name, 'rb'))) for img in images]
+    data = {
+        'topic': topic,
+        'aspect_ratio': aspect_ratio
+    }
+    response = requests.post(f"{BASE_URL}/generate_banner", data=data, files=files)
+    
+    # Close the files to prevent issues
+    for _, (_, file) in files:
+        file.close()
+    
+    if response.status_code == 200:
+        result = response.json()
+        image_path = result.get("image_path")
+        if image_path:
+            # Construct the full URL to access the generated image
+            full_image_url = f"{BASE_URL}{image_path}"
+            return full_image_url
+        else:
+            return "Error: Generated image not found."
+    else:
+        return "Error generating banner."
+
+# Function to preview uploaded images
+def preview_images(images):
+    return images
+
+# Gradio Interface
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    # Main Container for UI elements
+    with gr.Row():
+        # Left Column: Introduction, Banner Topic, Aspect Ratio, Upload Images, and Generate Button
+        with gr.Column(scale=1, min_width=300, elem_id="branding_section"):
+            gr.Markdown(
+                """
+                <div style="text-align: left;">
+                    <h1 style="font-size: 2.5rem; color: #ff6347; font-weight: bold;">Bigbasket Banner Generator</h1>
+                    <p style="font-size: 1.25rem; color: #333;">Create beautiful, impactful banners with ease! Simply upload images, enter the topic, and let our AI generate stunning banners for your projects.</p>
+                    <p style="font-size: 1.1rem; color: #555;">Powered by <b>Google Gemini</b> and <b>Google Imagen</b>.</p>
+                </div>
+                """, elem_id="intro_text"
+            )
+            # Moved Banner Topic Input Below "Powered by" Text
+            topic_input = gr.Textbox(label="📝 Banner Details (Product, Theme, Color, etc.)", placeholder="Enter the details for your banner", lines=2, elem_id="topic_input")
+            aspect_ratio_dropdown = gr.Dropdown(
+                label="📐 Select Aspect Ratio",
+                choices=["1:1", "9:16", "16:9", "4:3", "3:4"],
+                elem_id="aspect_ratio_input"
+            )
+            image_input = gr.Files(label="📸 Upload Images", file_types=["image"], type="filepath", elem_id="image_input")
+            generate_button = gr.Button("✨ Generate Banner", size="large", elem_id="generate_button")
+
+        # Right Column: Image Preview and Generated Banner
+        with gr.Column(scale=1, min_width=400, elem_id="interaction_section"):
+            image_preview = gr.Gallery(label="👀 Image Preview", elem_id="image_preview", show_label=False, height=300, preview=True)
+            generated_image = gr.Image(label="🎨 Generated Banner", type="filepath", interactive=False, elem_id="generated_image")
+
+            # Show uploaded images in the preview
+            image_input.change(
+                fn=preview_images,
+                inputs=image_input,
+                outputs=image_preview
+            )
+
+            # Button to Trigger Banner Generation and Display the Generated Image
+            generate_button.click(
+                fn=generate_banner,
+                inputs=[topic_input, image_input, aspect_ratio_dropdown],
+                outputs=generated_image
+            )
+
+    # Footer
+    gr.Markdown(
+        """
+        <div style="text-align: center; margin-top: 20px;">
+            <p style="font-size: 1.25rem; color: #888;">Made with ❤️ for creative designers</p>
+            <p style="font-size: 1rem; color: #888;">Powered by Google Gemini and Google Imagen</p>
+        </div>
+        """, elem_id="footer_text"
+    )
+
+# Custom CSS for improving the UI layout
+demo.css = """
+    #branding_section {
+        background-color: #f9f9f9;
+        padding: 20px;
+        border-radius: 15px;
+    }
+    #interaction_section {
+        padding: 20px;
+        background-color: #ffffff;
+        border-radius: 15px;
+        box-shadow: 0px 4px 8px rgba(0, 0, 0, 0.1);
+    }
+    #generate_button {
+        width: 100%;
+        margin-bottom: 15px;
+        padding: 15px;
+        font-size: 1.2rem;
+        font-weight: bold;
+        background-color: #ff6347;
+        color: white;
+        border: none;
+        border-radius: 10px;
+        cursor: pointer;
+    }
+    #generate_button:hover {
+        background-color: #e5533d;
+    }
+    #generated_image, #image_preview {
+        width: 100%;
+        border: none;
+        border-radius: 0px;
+        object-fit: cover;
+    }
+"""
+
+# Launch the demo
+demo.launch(share=True)

notebooks/__init__.py

@@ -0,0 +1 @@
+

requirements.txt

@@ -0,0 +1,7 @@
+configparser==7.1.0
+google-generativeai==0.8.1
+vertexai==1.66.0
+fastapi==0.115.0
+gradio==4.44.0
+uvicorn==0.31.0
+google-cloud-aiplatform==1.66.0