main.py

import os
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from io import BytesIO
import base64

# Import the required packages for the language model and prompts
from langchain_groq import ChatGroq
from langchain_core.prompts import PromptTemplate

# Import packages for image processing and Google GenAI client
from google import genai
from google.genai import types
from PIL import Image

# --- Global Initialization ---

# Get API keys from environment variables
LANGCHAIN_API_KEY = os.getenv("LANGCHAIN_API_KEY")
GOOGLE_API_KEY = os.getenv("GOOGLE_API_KEY")

if not LANGCHAIN_API_KEY or not GOOGLE_API_KEY:
    raise EnvironmentError("API keys for LANGCHAIN_API_KEY and GOOGLE_API_KEY must be set as environment variables.")

def get_llm(api_key: str):
    """
    Returns the ChatGroq LLM instance.
    """
    llm = ChatGroq(
        model="meta-llama/llama-4-scout-17b-16e-instruct",
        temperature=1,
        max_tokens=1024,
        api_key=api_key
    )
    return llm

# Initialize the LLM instance
llm = get_llm(LANGCHAIN_API_KEY)

# Define the prompt template for enhancing raw prompts.
prompt_template = PromptTemplate.from_template('''
You are a Prompt Enhancement AI Assistant specialized in crafting marketing posters that effectively communicate a brand's message and capture audience attention. Using the provided business description and the user's raw poster concept, generate a detailed, professional prompt optimized for AI-driven poster creation.

Enhance the prompt by including:
- always write generate image, dont use more text
- Visual Style: (e.g., flat design, isometric, photorealistic, illustrative)
- Color Scheme & Branding: (e.g., vibrant brand colors, pastel accents, monochrome palette)
- Typography: (e.g., bold sans-serif headline, elegant serif subtext)
- Composition & Layout: (e.g., focal point placement, use of negative space, layered elements)
- Imagery Elements: (e.g., product shots, icons, background patterns, textures)
- Lighting & Mood: (e.g., bright, dramatic shadows, soft ambient glow)
- Contextual Details: (e.g., logo placement, tagline integration, call-to-action placement)
- Call-to-Action: include a clear, concise CTA in the poster text (e.g., "Shop Now", "Learn More")

Business Description:
{Raw_Prompt}

Enhanced Poster Prompt:
''')

# Initialize the Google GenAI client.
client1 = genai.Client(api_key=GOOGLE_API_KEY)

# --- Request Models ---

class EnhancePromptRequest(BaseModel):
    Raw_Prompt: str

class GenerateImageRequest(BaseModel):
    # If both are provided, enhanced_prompt takes priority.
    Raw_Prompt: str = None
    enhanced_prompt: str = None

class UpdateImageRequest(BaseModel):
    text_instruction: str
    # Image encoded in base64
    image_base64: str

# --- FastAPI Initialization ---

app = FastAPI(title="Image Generation & Update API")

# 1. Root endpoint
@app.get("/")
async def root():
    return {"message": "Welcome to the Image Generation API!"}

# 2. Enhance Prompt endpoint
@app.post("/enhance-prompt")
async def enhance_prompt(request: EnhancePromptRequest):
    try:
        # Prepare the prompt using the template.
        formatted_prompt = prompt_template.invoke({"Raw_Prompt": request.Raw_Prompt})
        # Call the LLM to enhance the prompt.
        response = llm.invoke(formatted_prompt)
        # Assume the enhanced prompt is in the response.content.
        enhanced_prompt = response.content
        return {"raw_prompt": request.raw_prompt, "enhanced_prompt": enhanced_prompt}
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

# Helper function to generate image content using GenAI client.
def generate_image_from_prompt(image_prompt: str):
    response = client1.models.generate_content(
        model="gemini-2.0-flash-exp-image-generation",
        contents=image_prompt,
        config=types.GenerateContentConfig(
            response_modalities=['Text', 'Image']
        )
    )
    # Prepare a dict to hold results.
    result = {"text": None, "image_base64": None}
    # Process the response parts.
    for part in response.candidates[0].content.parts:
        if part.text is not None:
            result["text"] = part.text
        elif part.inline_data is not None:
            # Open the image from bytes.
            image = Image.open(BytesIO(part.inline_data.data))
            # Convert image to bytes.
            buffered = BytesIO()
            image.save(buffered, format="PNG")
            # Encode to base64.
            img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
            result["image_base64"] = img_str
    return result

# 3. Generate Image endpoint
@app.post("/generate-poster")
async def generate_image(request: GenerateImageRequest):
    try:
        # Decide which prompt to use.
        if request.enhanced_prompt:
            image_prompt = request.enhanced_prompt
        elif request.raw_prompt:
            image_prompt = request.Raw_Prompt
        else:
            raise HTTPException(status_code=400, detail="Either raw_prompt or enhanced_prompt must be provided.")
        
        result = generate_image_from_prompt(image_prompt)
        return result
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

# 4. Update Image endpoint
@app.post("/update-poster")
async def update_image(request: UpdateImageRequest):
    try:
        # Decode the incoming base64 image.
        image_bytes = base64.b64decode(request.image_base64)
        image = Image.open(BytesIO(image_bytes))
        
        # The contents for the update API include the text instruction and the current image.
        response = client1.models.generate_content(
            model="gemini-2.0-flash-exp-image-generation",
            contents=[request.text_instruction, image],
            config=types.GenerateContentConfig(
                response_modalities=['Text', 'Image']
            )
        )
        updated_result = {"text": None, "updated_image_base64": None}
        for part in response.candidates[0].content.parts:
            if part.text is not None:
                updated_result["text"] = part.text
            elif part.inline_data is not None:
                updated_image = Image.open(BytesIO(part.inline_data.data))
                buffered = BytesIO()
                updated_image.save(buffered, format="PNG")
                updated_img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
                updated_result["updated_image_base64"] = updated_img_str
        return updated_result
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

# --- Run the app ---
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=8000)