main.py

from fastapi import FastAPI, HTTPException, BackgroundTasks, Depends
from pydantic import BaseModel
from fastapi.staticfiles import StaticFiles
from fastapi.concurrency import run_in_threadpool
from fastapi.middleware.cors import CORSMiddleware
import uuid
import time
import tempfile
from concurrent.futures import ThreadPoolExecutor
from pymongo import MongoClient
from urllib.parse import quote_plus
from langchain_groq import ChatGroq
from aura_sr import AuraSR
from io import BytesIO
from PIL import Image
import requests
import os
import logging
from dotenv import load_dotenv

# Load environment variables
load_dotenv()

# Validate environment variables
assert os.getenv('MONGO_USER') and os.getenv('MONGO_PASSWORD') and os.getenv('MONGO_HOST'), "MongoDB credentials missing!"
assert os.getenv('LLM_API_KEY'), "LLM API Key missing!"
assert os.getenv('BFL_API_KEY'), "BFL API Key missing!"

# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)

# Set the Hugging Face cache directory to a writable location
os.environ['HF_HOME'] = '/tmp/huggingface_cache'

app = FastAPI()

# Middleware for CORS
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

# Globals
executor = ThreadPoolExecutor(max_workers=5)
llm = None
mongo_client = MongoClient(f"mongodb+srv://{os.getenv('MONGO_USER')}:{quote_plus(os.getenv('MONGO_PASSWORD'))}@{os.getenv('MONGO_HOST')}/")
db = mongo_client["Flux"]
collection = db["chat_histories"]
chat_sessions = {}

# Temporary directory for storing images
image_storage_dir = tempfile.mkdtemp()
app.mount("/images", StaticFiles(directory=image_storage_dir), name="images")

# Initialize AuraSR during startup
aura_sr = None

@app.on_event("startup")
async def startup():
    global llm, aura_sr
    try:
        llm = ChatGroq(
            model="llama-3.3-70b-versatile",
            temperature=0.7,
            max_tokens=1024,
            api_key=os.getenv('LLM_API_KEY'),
        )
        aura_sr = AuraSR.from_pretrained("fal/AuraSR-v2")
    except Exception as e:
        logger.error(f"Error initializing models: {e}")

@app.on_event("shutdown")
def shutdown():
    mongo_client.close()
    executor.shutdown()

# Pydantic models
class ImageRequest(BaseModel):
    subject: str
    style: str
    color_theme: str
    elements: str
    color_mode: str
    lighting_conditions: str
    framing_style: str
    material_details: str
    text: str
    background_details: str
    user_prompt: str
    chat_id: str

class UpscaleRequest(BaseModel):
    image_url: str

# Helper functions
def generate_chat_id():
    chat_id = str(uuid.uuid4())
    chat_sessions[chat_id] = collection
    return chat_id

def get_chat_history(chat_id):
    messages = collection.find({"session_id": chat_id})
    return "\n".join(
        f"User: {msg['content']}" if msg['role'] == "user" else f"AI: {msg['content']}"
        for msg in messages
    )

def save_image_locally(image, filename):
    filepath = os.path.join(image_storage_dir, filename)
    image.save(filepath, format="PNG")
    return filepath


def make_request_with_retries(url, headers, payload, retries=3, delay=2):
    """
    Makes an HTTP POST request with retries in case of failure.
    :param url: The URL for the request.
    :param headers: Headers to include in the request.
    :param payload: Payload to include in the request.
    :param retries: Number of retries on failure.
    :param delay: Delay between retries.
    :return: Response JSON from the server.
    """
    for attempt in range(retries):
        try:
            with requests.Session() as session:
                response = session.post(url, headers=headers, json=payload, timeout=30)
                response.raise_for_status()
                return response.json()
        except requests.exceptions.RequestException as e:
            if attempt < retries - 1:
                time.sleep(delay)
                continue
            else:
                raise HTTPException(status_code=500, detail=f"Request failed after {retries} attempts: {str(e)}")


def fetch_image(url):
    try:
        with requests.Session() as session:
            response = session.get(url, timeout=30)
            response.raise_for_status()
            return Image.open(BytesIO(response.content))
    except Exception as e:
        raise HTTPException(status_code=400, detail=f"Error fetching image: {str(e)}")

def poll_for_image_result(request_id, headers):
    timeout = 60
    start_time = time.time()

    while time.time() - start_time < timeout:
        time.sleep(0.5)
        with requests.Session() as session:
            result = session.get(
                "https://api.bfl.ml/v1/get_result",
                headers=headers,
                params={"id": request_id},
                timeout=10
            ).json()

        if result["status"] == "Ready":
            return result["result"].get("sample")
        elif result["status"] == "Error":
            raise HTTPException(status_code=500, detail=f"Image generation failed: {result.get('error', 'Unknown error')}")

    raise HTTPException(status_code=500, detail="Image generation timed out.")

@app.post("/new-chat", response_model=dict)
async def new_chat():
    chat_id = generate_chat_id()
    return {"chat_id": chat_id}

@app.post("/generate-image", response_model=dict)
async def generate_image(request: ImageRequest):
    chat_history = get_chat_history(request.chat_id)
    prompt = f"""
    You are a professional assistant responsible for crafting a clear and visually compelling prompt for an image generation model. Your task is to generate a high-quality prompt for creating both the **main subject** and the **background** of the image.

    Image Specifications:
    - **Subject**: Focus on **{request.subject}**, highlighting its defining features, expressions, and textures.
    - **Style**: Emphasize the **{request.style}**, capturing its key characteristics.
    - **Background**: Create a background with **{request.background_details}** that complements and enhances the subject. Ensure it aligns with the color theme and overall composition.
    - **Camera and Lighting**:
      - Lighting: Apply **{request.lighting_conditions}**, emphasizing depth, highlights, and shadows to accentuate the subject and harmonize the background.
    - **Framing**: Use a **{request.framing_style}** to enhance the composition around both the subject and the background.
    - **Materials**: Highlight textures like **{request.material_details}**, with realistic details and natural imperfections on the subject and background.
    - **Key Elements**: Include **{request.elements}** to enrich the subject’s details and add cohesive elements to the background.
    - **Color Theme**: Follow the **{request.color_theme}** to set the mood and tone for the entire scene.
    - Negative Prompt: Avoid grainy, blurry, or deformed outputs.
    - **Text to Include in Image**: Clearly display the text **"{request.text}"** as part of the composition (e.g., on a card, badge, or banner) attached to the subject in a realistic and contextually appropriate way.
    """

    refined_prompt = llm.invoke(prompt).content.strip()
    collection.insert_one({"session_id": request.chat_id, "role": "user", "content": request.user_prompt})
    collection.insert_one({"session_id": request.chat_id, "role": "ai", "content": refined_prompt})

    headers = {
        "accept": "application/json",
        "x-key": os.getenv('BFL_API_KEY'),
        "Content-Type": "application/json"
    }

    payload = {
        "prompt": refined_prompt,
        "width": 1024,
        "height": 1024,
        "guidance_scale": 1,
        "num_inference_steps": 50,
        "max_sequence_length": 512,
    }

    response = make_request_with_retries("https://api.bfl.ml/v1/flux-pro-1.1", headers, payload)

    if "id" not in response:
        raise HTTPException(status_code=500, detail="Error generating image: ID missing from response")

    image_url = poll_for_image_result(response["id"], headers)
    image = fetch_image(image_url)
    filename = f"generated_{uuid.uuid4()}.png"
    filepath = save_image_locally(image, filename)

    return {
        "status": "Image generated successfully",
        "file_path": filepath,
        "file_url": f"/images/{filename}",
    }

@app.post("/upscale-image", response_model=dict)
async def upscale_image(request: UpscaleRequest):
    if aura_sr is None:
        raise HTTPException(status_code=500, detail="Upscaling model not initialized.")

    img = await run_in_threadpool(fetch_image, request.image_url)

    def perform_upscaling():
        upscaled_image = aura_sr.upscale_4x_overlapped(img)
        filename = f"upscaled_{uuid.uuid4()}.png"
        return save_image_locally(upscaled_image, filename)

    future = executor.submit(perform_upscaling)
    filepath = await run_in_threadpool(lambda: future.result())

    return {
        "status": "Upscaling successful",
        "file_path": filepath,
        "file_url": f"/images/{os.path.basename(filepath)}",
    }

@app.get("/")
async def root():
    return {"message": "API is up and running!"}