Create media_clicks_app.py

media_clicks_app.py

@@ -0,0 +1,1300 @@
+#####################FASTAPI___________________##############
+import os
+os.environ["OMP_NUM_THREADS"] = "1"
+import shutil
+import uuid
+import cv2
+import numpy as np
+import threading
+import asyncio
+import subprocess
+import logging
+import tempfile
+import sys
+import time
+from datetime import datetime,timedelta
+import tempfile
+import insightface
+from insightface.app import FaceAnalysis
+from huggingface_hub import hf_hub_download
+from fastapi import FastAPI, UploadFile, File, HTTPException, Response, Depends, Security, Form
+from fastapi.responses import RedirectResponse
+from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
+from motor.motor_asyncio import AsyncIOMotorClient
+from bson import ObjectId
+from bson.errors import InvalidId
+import httpx
+import uvicorn
+from PIL import Image
+import io
+import requests
+# DigitalOcean Spaces
+import boto3
+from botocore.client import Config
+from typing import Optional
+
+# --------------------- Logging ---------------------
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger(__name__)
+
+# --------------------- Secrets & Paths ---------------------
+REPO_ID = "HariLogicgo/face_swap_models"
+MODELS_DIR = "./models"
+os.makedirs(MODELS_DIR, exist_ok=True)
+
+HF_TOKEN = os.getenv("HF_TOKEN")
+API_SECRET_TOKEN = os.getenv("API_SECRET_TOKEN")
+
+DO_SPACES_REGION = os.getenv("DO_SPACES_REGION", "blr1")
+DO_SPACES_ENDPOINT = f"https://{DO_SPACES_REGION}.digitaloceanspaces.com"
+DO_SPACES_KEY = os.getenv("DO_SPACES_KEY")
+DO_SPACES_SECRET = os.getenv("DO_SPACES_SECRET")
+DO_SPACES_BUCKET = os.getenv("DO_SPACES_BUCKET")
+
+# NEW admin DB (with error handling for missing env vars)
+ADMIN_MONGO_URL = os.getenv("ADMIN_MONGO_URL")
+admin_client = None
+admin_db = None
+subcategories_col = None
+if ADMIN_MONGO_URL:
+    try:
+        admin_client = AsyncIOMotorClient(ADMIN_MONGO_URL)
+        admin_db = admin_client.adminPanel
+        subcategories_col = admin_db.subcategories
+    except Exception as e:
+        logger.warning(f"MongoDB admin connection failed (optional): {e}")
+
+# Collage Maker DB (optional)
+COLLAGE_MAKER_DB_URL = os.getenv("COLLAGE_MAKER_DB_URL")
+collage_maker_client = None
+collage_maker_db = None
+collage_subcategories_col = None
+if COLLAGE_MAKER_DB_URL:
+    try:
+        collage_maker_client = AsyncIOMotorClient(COLLAGE_MAKER_DB_URL)
+        collage_maker_db = collage_maker_client.adminPanel
+        collage_subcategories_col = collage_maker_db.subcategories
+    except Exception as e:
+        logger.warning(f"MongoDB collage-maker connection failed (optional): {e}")
+        
+# AI Enhancer DB (optional)
+
+AI_ENHANCER_DB_URL = os.getenv("AI_ENHANCER_DB_URL")
+ai_enhancer_client = None
+ai_enhancer_db = None
+ai_enhancer_subcategories_col = None
+
+if AI_ENHANCER_DB_URL:
+    try:
+        ai_enhancer_client = AsyncIOMotorClient(AI_ENHANCER_DB_URL)
+        ai_enhancer_db = ai_enhancer_client.test   # 🔴 test database
+        ai_enhancer_subcategories_col = ai_enhancer_db.subcategories
+    except Exception as e:
+        logger.warning(f"MongoDB ai-enhancer connection failed (optional): {e}")
+
+
+# OLD logs DB
+# MONGODB_URL = os.getenv("MONGODB_URL")
+# client = None
+# database = None
+FACESWAP_LOGS_MONGO_URL = os.getenv("MONGODB_URL")
+
+logs_client = None
+logs_collection = None
+# --------------------- Download Models ---------------------
+def download_models():
+    try:
+        logger.info("Downloading models...")
+        inswapper_path = hf_hub_download(
+            repo_id=REPO_ID,
+            filename="models/inswapper_128.onnx",
+            repo_type="model",
+            local_dir=MODELS_DIR,
+            token=HF_TOKEN
+        )
+
+        buffalo_files = ["1k3d68.onnx", "2d106det.onnx", "genderage.onnx", "det_10g.onnx", "w600k_r50.onnx"]
+        for f in buffalo_files:
+            hf_hub_download(
+                repo_id=REPO_ID,
+                filename=f"models/buffalo_l/" + f,
+                repo_type="model",
+                local_dir=MODELS_DIR,
+                token=HF_TOKEN
+            )
+
+        logger.info("Models downloaded successfully.")
+        return inswapper_path
+    except Exception as e:
+        logger.error(f"Model download failed: {e}")
+        raise
+
+try:
+    inswapper_path = download_models()
+    
+    # --------------------- Face Analysis + Swapper ---------------------
+    providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
+    face_analysis_app = FaceAnalysis(name="buffalo_l", root=MODELS_DIR, providers=providers)
+    face_analysis_app.prepare(ctx_id=0, det_size=(640, 640))
+    swapper = insightface.model_zoo.get_model(inswapper_path, providers=providers)
+    logger.info("Face analysis models loaded successfully")
+except Exception as e:
+    logger.error(f"Failed to initialize face analysis models: {e}")
+    # Set defaults to prevent crash
+    inswapper_path = None
+    face_analysis_app = None
+    swapper = None
+
+# --------------------- CodeFormer ---------------------
+CODEFORMER_PATH = "CodeFormer/inference_codeformer.py"
+
+def ensure_codeformer():
+    """
+    Ensure CodeFormer's local basicsr + facelib are importable and
+    pretrained weights are downloaded. No setup.py needed — we use
+    sys.path / PYTHONPATH instead.
+    """
+    try:
+        if not os.path.exists("CodeFormer"):
+            logger.info("CodeFormer not found, cloning repository...")
+            subprocess.run("git clone https://github.com/sczhou/CodeFormer.git", shell=True, check=True)
+            subprocess.run("pip install -r CodeFormer/requirements.txt", shell=True, check=False)
+        
+        # Add CodeFormer root to sys.path so `import basicsr` and
+        # `import facelib` resolve to the local (compatible) versions
+        # instead of the broken PyPI basicsr==1.4.2.
+        codeformer_root = os.path.join(os.getcwd(), "CodeFormer")
+        if codeformer_root not in sys.path:
+            sys.path.insert(0, codeformer_root)
+            logger.info(f"Added {codeformer_root} to sys.path for local basicsr/facelib")
+        
+        # NOTE: We do NOT need the PyPI 'realesrgan' package.
+        # Both in-process and subprocess paths use CodeFormer's local
+        # basicsr.utils.realesrgan_utils.RealESRGANer instead.
+        # Installing PyPI realesrgan at runtime would re-install the
+        # broken basicsr==1.4.2 and break everything.
+        
+        # Download pretrained weights if not already present
+        if os.path.exists("CodeFormer"):
+            try:
+                subprocess.run("python CodeFormer/scripts/download_pretrained_models.py facelib", shell=True, check=False, timeout=300)
+            except (subprocess.TimeoutExpired, subprocess.CalledProcessError):
+                logger.warning("Failed to download facelib models (optional)")
+            try:
+                subprocess.run("python CodeFormer/scripts/download_pretrained_models.py CodeFormer", shell=True, check=False, timeout=300)
+            except (subprocess.TimeoutExpired, subprocess.CalledProcessError):
+                logger.warning("Failed to download CodeFormer models (optional)")
+    except Exception as e:
+        logger.error(f"CodeFormer setup failed: {e}")
+        logger.warning("Continuing without CodeFormer features...")
+
+ensure_codeformer()
+
+# --------------------- In-Process CodeFormer (No Subprocess!) ---------------------
+# Load CodeFormer models ONCE at startup instead of spawning a new Python process per request.
+# This eliminates 15-40s of model loading overhead per request.
+
+codeformer_net = None
+codeformer_upsampler = None
+codeformer_face_helper = None
+codeformer_device = None
+
+def init_codeformer_in_process():
+    """Load CodeFormer models once into memory for fast per-request inference."""
+    global codeformer_net, codeformer_upsampler, codeformer_face_helper, codeformer_device
+    try:
+        import torch
+        from torchvision.transforms.functional import normalize as torch_normalize
+        
+        # Add CodeFormer to Python path
+        codeformer_root = os.path.join(os.getcwd(), "CodeFormer")
+        if codeformer_root not in sys.path:
+            sys.path.insert(0, codeformer_root)
+        
+        from basicsr.utils.registry import ARCH_REGISTRY
+        from basicsr.utils.download_util import load_file_from_url
+        from facelib.utils.face_restoration_helper import FaceRestoreHelper
+        
+        codeformer_device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
+        logger.info(f"Initializing CodeFormer on device: {codeformer_device}")
+        
+        # 1) Load CodeFormer network
+        net = ARCH_REGISTRY.get('CodeFormer')(
+            dim_embd=512, codebook_size=1024, n_head=8, n_layers=9,
+            connect_list=['32', '64', '128', '256']
+        ).to(codeformer_device)
+        
+        ckpt_path = load_file_from_url(
+            url='https://github.com/sczhou/CodeFormer/releases/download/v0.1.0/codeformer.pth',
+            model_dir='weights/CodeFormer', progress=True, file_name=None
+        )
+        checkpoint = torch.load(ckpt_path, map_location=codeformer_device)['params_ema']
+        net.load_state_dict(checkpoint)
+        net.eval()
+        codeformer_net = net
+        
+        # 2) RealESRGAN upsampler — SKIPPED for face swap
+        # Background/face upsampling is the #1 bottleneck (~20s per image).
+        # For face swap we only need CodeFormer face restoration, not super-resolution.
+        # The upsampler is kept as None; we no longer download the 64MB model at startup.
+        codeformer_upsampler = None
+        
+        # 3) Create FaceRestoreHelper (reused per request)
+        # NOTE: local CodeFormer uses "upscale_factor" (not "upscale")
+        # upscale_factor=1 → keep original resolution (no 2x upscale needed for face swap)
+        codeformer_face_helper = FaceRestoreHelper(
+            upscale_factor=1,
+            face_size=512,
+            crop_ratio=(1, 1),
+            det_model='retinaface_resnet50',
+            save_ext='png',
+            use_parse=True,
+            device=codeformer_device
+        )
+        
+        logger.info("✅ CodeFormer models loaded in-process successfully!")
+        return True
+    except Exception as e:
+        logger.error(f"Failed to load CodeFormer in-process: {e}")
+        logger.warning("CodeFormer enhancement will be unavailable.")
+        return False
+
+# Try to load CodeFormer models in-process
+_codeformer_available = init_codeformer_in_process()
+# --------------------- FastAPI ---------------------
+fastapi_app = FastAPI()
+
+@fastapi_app.on_event("startup")
+async def startup_db():
+    global logs_client, logs_collection
+
+    if FACESWAP_LOGS_MONGO_URL:
+        try:
+            logger.info("Initializing NEW logs DB → logs/faceswap-new")
+
+            logs_client = AsyncIOMotorClient(FACESWAP_LOGS_MONGO_URL)
+
+            # Force database and collection
+            logs_db = logs_client["logs"]
+            logs_collection = logs_db["faceswap-new"]
+
+            logger.info("MongoDB logs collection ready")
+        except Exception as e:
+            logger.warning(f"Logs MongoDB connection failed: {e}")
+            logs_client = None
+            logs_collection = None
+    else:
+        logger.warning("FACESWAP_LOGS_MONGO_URL not set")
+@fastapi_app.on_event("shutdown")
+async def shutdown_db():
+    # global client, admin_client, collage_maker_client
+    global logs_client, admin_client, collage_maker_client
+    # if client is not None:
+    #     client.close()
+    if logs_client is not None:
+        logs_client.close()
+        logger.info("MongoDB connection closed")
+    if admin_client is not None:
+        admin_client.close()
+        logger.info("Admin MongoDB connection closed")
+    if collage_maker_client is not None:
+        collage_maker_client.close()
+        logger.info("Collage Maker MongoDB connection closed")
+
+# --------------------- Auth ---------------------
+security = HTTPBearer()
+
+def verify_token(credentials: HTTPAuthorizationCredentials = Security(security)):
+    if credentials.credentials != API_SECRET_TOKEN:
+        raise HTTPException(status_code=401, detail="Invalid or missing token")
+    return credentials.credentials
+
+# --------------------- DB Selector ---------------------
+def get_app_db_collections(appname: Optional[str] = None):
+    """
+    Returns subcategories_collection based on appname.
+    """
+
+    if appname:
+        app = appname.strip().lower()
+
+        if app == "collage-maker":
+            if collage_subcategories_col is not None:
+                return collage_subcategories_col
+            logger.warning("Collage-maker DB not configured, falling back to admin")
+
+        elif app == "ai-enhancer":
+            if ai_enhancer_subcategories_col is not None:
+                return ai_enhancer_subcategories_col
+            logger.warning("AI-Enhancer DB not configured, falling back to admin")
+
+    # default fallback
+    return subcategories_col
+
+
+
+# --------------------- Logging API Hits ---------------------
+async def log_faceswap_hit(token: str, status: str = "success"):
+    global database
+    if database is None:
+        return
+    await database.api_logs.insert_one({
+        "token": token,
+        "endpoint": "/faceswap",
+        "status": status,
+        "timestamp": datetime.utcnow()
+    })
+
+# --------------------- Face Swap Pipeline ---------------------
+swap_lock = threading.Lock()
+
+def enhance_image_with_codeformer(rgb_img, temp_dir=None, w=0.7):
+    """
+    Enhance face image using CodeFormer.
+    Uses in-process models (fast) if available, falls back to subprocess (slow).
+    """
+    global codeformer_net, codeformer_upsampler, codeformer_face_helper, codeformer_device
+    
+    t0 = time.time()
+    
+    # ── FAST PATH: In-process CodeFormer (no subprocess!) ──
+    if codeformer_net is not None and codeformer_face_helper is not None:
+        import torch
+        from torchvision.transforms.functional import normalize as torch_normalize
+        from basicsr.utils import img2tensor, tensor2img
+        from facelib.utils.misc import is_gray
+        
+        bgr_img = cv2.cvtColor(rgb_img, cv2.COLOR_RGB2BGR)
+        
+        # Reset face helper state
+        codeformer_face_helper.clean_all()
+        codeformer_face_helper.read_image(bgr_img)
+        
+        num_faces = codeformer_face_helper.get_face_landmarks_5(
+            only_center_face=False, resize=640, eye_dist_threshold=5
+        )
+        logger.info(f"[CodeFormer] Detected {num_faces} faces in {time.time()-t0:.2f}s")
+        
+        codeformer_face_helper.align_warp_face()
+        
+        # Enhance each cropped face with CodeFormer neural net
+        t_faces = time.time()
+        for idx, cropped_face in enumerate(codeformer_face_helper.cropped_faces):
+            cropped_face_t = img2tensor(cropped_face / 255., bgr2rgb=True, float32=True)
+            torch_normalize(cropped_face_t, (0.5, 0.5, 0.5), (0.5, 0.5, 0.5), inplace=True)
+            cropped_face_t = cropped_face_t.unsqueeze(0).to(codeformer_device)
+            
+            try:
+                with torch.no_grad():
+                    output = codeformer_net(cropped_face_t, w=w, adain=True)[0]
+                    restored_face = tensor2img(output, rgb2bgr=True, min_max=(-1, 1))
+                del output
+                torch.cuda.empty_cache()
+            except Exception as e:
+                logger.warning(f"[CodeFormer] Face {idx} inference failed: {e}")
+                restored_face = tensor2img(cropped_face_t, rgb2bgr=True, min_max=(-1, 1))
+            
+            restored_face = restored_face.astype('uint8')
+            codeformer_face_helper.add_restored_face(restored_face, cropped_face)
+        logger.info(f"[CodeFormer] Face restoration ({num_faces} faces): {time.time()-t_faces:.2f}s")
+        
+        # Paste restored faces back onto original image
+        # NOTE: We skip RealESRGAN background/face upsampling — it's the #1 bottleneck
+        # (~20s) and unnecessary for face swap. We only need CodeFormer face restoration.
+        t_paste = time.time()
+        codeformer_face_helper.get_inverse_affine(None)
+        restored_img = codeformer_face_helper.paste_faces_to_input_image(
+            upsample_img=None, draw_box=False
+        )
+        logger.info(f"[CodeFormer] Paste back: {time.time()-t_paste:.2f}s")
+        
+        logger.info(f"[CodeFormer] In-process enhancement done in {time.time()-t0:.2f}s")
+        return cv2.cvtColor(restored_img, cv2.COLOR_BGR2RGB)
+    
+    # ── SLOW FALLBACK: Subprocess CodeFormer (with timeout!) ──
+    logger.warning("[CodeFormer] In-process models unavailable, falling back to subprocess")
+    if temp_dir is None:
+        temp_dir = os.path.join(tempfile.gettempdir(), f"enhance_{uuid.uuid4().hex[:8]}")
+    os.makedirs(temp_dir, exist_ok=True)
+
+    input_path = os.path.join(temp_dir, "input.jpg")
+    cv2.imwrite(input_path, cv2.cvtColor(rgb_img, cv2.COLOR_RGB2BGR))
+
+    python_cmd = sys.executable if sys.executable else "python3"
+    cmd = (
+        f"{python_cmd} {CODEFORMER_PATH} "
+        f"-w {w} "
+        f"--input_path {input_path} "
+        f"--output_path {temp_dir} "
+        f"--bg_upsampler None "
+        f"--upscale 1"
+    )
+
+    result = subprocess.run(cmd, shell=True, capture_output=True, text=True, timeout=120)
+    if result.returncode != 0:
+        raise RuntimeError(result.stderr)
+
+    final_dir = os.path.join(temp_dir, "final_results")
+    files = [f for f in os.listdir(final_dir) if f.endswith(".png")]
+    if not files:
+        raise RuntimeError("No enhanced output")
+
+    final_path = os.path.join(final_dir, files[0])
+    enhanced = cv2.imread(final_path)
+    logger.info(f"[CodeFormer] Subprocess enhancement done in {time.time()-t0:.2f}s")
+    return cv2.cvtColor(enhanced, cv2.COLOR_BGR2RGB)
+    
+def multi_face_swap(src_img, tgt_img):
+    pipeline_start = time.time()
+    src_bgr = cv2.cvtColor(src_img, cv2.COLOR_RGB2BGR)
+    tgt_bgr = cv2.cvtColor(tgt_img, cv2.COLOR_RGB2BGR)
+
+    t0 = time.time()
+    src_faces = face_analysis_app.get(src_bgr)
+    tgt_faces = face_analysis_app.get(tgt_bgr)
+    logger.info(f"[Pipeline] Multi-face detection: {time.time()-t0:.2f}s")
+
+    if not src_faces or not tgt_faces:
+        raise ValueError("No faces detected")
+
+    def face_sort_key(face):
+        x1, y1, x2, y2 = face.bbox
+        area = (x2 - x1) * (y2 - y1)
+        cx = (x1 + x2) / 2
+        return (-area, cx)
+
+    src_male = sorted([f for f in src_faces if f.gender == 1], key=face_sort_key)
+    src_female = sorted([f for f in src_faces if f.gender == 0], key=face_sort_key)
+    tgt_male = sorted([f for f in tgt_faces if f.gender == 1], key=face_sort_key)
+    tgt_female = sorted([f for f in tgt_faces if f.gender == 0], key=face_sort_key)
+
+    pairs = []
+    for s, t in zip(src_male, tgt_male):
+        pairs.append((s, t))
+    for s, t in zip(src_female, tgt_female):
+        pairs.append((s, t))
+
+    if not pairs:
+        src_faces = sorted(src_faces, key=face_sort_key)
+        tgt_faces = sorted(tgt_faces, key=face_sort_key)
+        pairs = list(zip(src_faces, tgt_faces))
+
+    t0 = time.time()
+    result_img = tgt_bgr.copy()
+    for src_face, _ in pairs:
+        if face_analysis_app is None:
+            raise ValueError("Face analysis models not initialized.")
+        current_faces = sorted(face_analysis_app.get(result_img), key=face_sort_key)
+        candidates = [f for f in current_faces if f.gender == src_face.gender] or current_faces
+        target_face = candidates[0]
+
+        if swapper is None:
+            raise ValueError("Face swap models not initialized.")
+        result_img = swapper.get(result_img, target_face, src_face, paste_back=True)
+    logger.info(f"[Pipeline] Multi-face swap ({len(pairs)} pairs): {time.time()-t0:.2f}s")
+
+    logger.info(f"[Pipeline] TOTAL multi_face_swap: {time.time()-pipeline_start:.2f}s")
+    return cv2.cvtColor(result_img, cv2.COLOR_BGR2RGB)
+
+
+
+def face_swap_and_enhance(src_img, tgt_img, temp_dir=None):
+    try:
+        with swap_lock:
+            pipeline_start = time.time()
+            
+            if temp_dir is None:
+                temp_dir = os.path.join(tempfile.gettempdir(), f"faceswap_work_{uuid.uuid4().hex[:8]}")
+            if os.path.exists(temp_dir):
+                shutil.rmtree(temp_dir)
+            os.makedirs(temp_dir, exist_ok=True)
+
+            if face_analysis_app is None:
+                return None, None, "❌ Face analysis models not initialized."
+            if swapper is None:
+                return None, None, "❌ Face swap models not initialized."
+
+            src_bgr = cv2.cvtColor(src_img, cv2.COLOR_RGB2BGR)
+            tgt_bgr = cv2.cvtColor(tgt_img, cv2.COLOR_RGB2BGR)
+
+            t0 = time.time()
+            src_faces = face_analysis_app.get(src_bgr)
+            tgt_faces = face_analysis_app.get(tgt_bgr)
+            logger.info(f"[Pipeline] Face detection: {time.time()-t0:.2f}s")
+            
+            if not src_faces or not tgt_faces:
+                return None, None, "❌ Face not detected in one of the images"
+
+            t0 = time.time()
+            swapped_bgr = swapper.get(tgt_bgr, tgt_faces[0], src_faces[0])
+            logger.info(f"[Pipeline] Face swap: {time.time()-t0:.2f}s")
+            
+            if swapped_bgr is None:
+                return None, None, "❌ Face swap failed"
+
+            # Use in-process CodeFormer enhancement (fast path)
+            t0 = time.time()
+            swapped_rgb = cv2.cvtColor(swapped_bgr, cv2.COLOR_BGR2RGB)
+            try:
+                enhanced_rgb = enhance_image_with_codeformer(swapped_rgb)
+                enhanced_bgr = cv2.cvtColor(enhanced_rgb, cv2.COLOR_RGB2BGR)
+            except Exception as e:
+                logger.error(f"[Pipeline] CodeFormer failed, using raw swap: {e}")
+                enhanced_bgr = swapped_bgr
+            logger.info(f"[Pipeline] Enhancement: {time.time()-t0:.2f}s")
+
+            final_path = os.path.join(temp_dir, f"result_{uuid.uuid4().hex[:8]}.png")
+            cv2.imwrite(final_path, enhanced_bgr)
+            
+            final_img = cv2.cvtColor(enhanced_bgr, cv2.COLOR_BGR2RGB)
+            
+            logger.info(f"[Pipeline] TOTAL face_swap_and_enhance: {time.time()-pipeline_start:.2f}s")
+            return final_img, final_path, ""
+
+    except Exception as e:
+        return None, None, f"❌ Error: {str(e)}"
+        
+def compress_image(
+    image_bytes: bytes,
+    max_size=(1280, 1280),   # max width/height
+    quality=75               # JPEG quality (60–80 is ideal)
+) -> bytes:
+    """
+    Compress image by resizing and lowering quality.
+    Returns compressed image bytes.
+    """
+    img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+
+    # Resize while maintaining aspect ratio
+    img.thumbnail(max_size, Image.LANCZOS)
+
+    output = io.BytesIO()
+    img.save(
+        output,
+        format="JPEG",
+        quality=quality,
+        optimize=True,
+        progressive=True
+    )
+
+    return output.getvalue()
+
+# --------------------- DigitalOcean Spaces Helper ---------------------
+def get_spaces_client():
+    session = boto3.session.Session()
+    client = session.client(
+        's3',
+        region_name=DO_SPACES_REGION,
+        endpoint_url=DO_SPACES_ENDPOINT,
+        aws_access_key_id=DO_SPACES_KEY,
+        aws_secret_access_key=DO_SPACES_SECRET,
+        config=Config(signature_version='s3v4')
+    )
+    return client
+
+def upload_to_spaces(file_bytes, key, content_type="image/png"):
+    client = get_spaces_client()
+    client.put_object(Bucket=DO_SPACES_BUCKET, Key=key, Body=file_bytes, ContentType=content_type, ACL='public-read')
+    return f"{DO_SPACES_ENDPOINT}/{DO_SPACES_BUCKET}/{key}"
+
+def download_from_spaces(key):
+    client = get_spaces_client()
+    obj = client.get_object(Bucket=DO_SPACES_BUCKET, Key=key)
+    return obj['Body'].read()
+
+def mandatory_enhancement(rgb_img):
+    """
+    Always runs CodeFormer on the final image.
+    Fail-safe: returns original if enhancement fails.
+    """
+    try:
+        return enhance_image_with_codeformer(rgb_img)
+    except Exception as e:
+        logger.error(f"CodeFormer failed, returning original: {e}")
+        return rgb_img
+
+# --------------------- API Endpoints ---------------------
+@fastapi_app.get("/")
+async def root():
+    """Root endpoint"""
+    return {
+        "success": True,
+        "message": "FaceSwap API",
+        "data": {
+            "version": "1.0.0",
+            "Product Name":"Beauty Camera - GlowCam AI Studio",
+            "Released By" : "LogicGo Infotech"
+        }
+    }
+@fastapi_app.get("/health")
+async def health():
+    return {"status": "healthy"}
+
+@fastapi_app.get("/test-admin-db")
+async def test_admin_db():
+    try:
+        doc = await admin_db.list_collection_names()
+        return {"ok": True, "collections": doc}
+    except Exception as e:
+        return {"ok": False, "error": str(e), "url": ADMIN_MONGO_URL}
+
+@fastapi_app.post("/face-swap", dependencies=[Depends(verify_token)])
+async def face_swap_api(
+    source: UploadFile = File(...),
+    image2: Optional[UploadFile] = File(None),
+    target_category_id: str = Form(None),
+    new_category_id: str = Form(None),
+    user_id: Optional[str] = Form(None),
+    appname: Optional[str] = Form(None),
+    credentials: HTTPAuthorizationCredentials = Security(security)
+):
+    start_time = datetime.utcnow()
+
+    try:
+        # ------------------------------------------------------------------
+        # VALIDATION
+        # ------------------------------------------------------------------
+        # --------------------------------------------------------------
+        # BACKWARD COMPATIBILITY FOR OLD ANDROID VERSIONS
+        # --------------------------------------------------------------
+        if target_category_id == "":
+            target_category_id = None
+        
+        if new_category_id == "":
+            new_category_id = None
+        
+        if user_id == "":
+            user_id = None
+
+        subcategories_collection = get_app_db_collections(appname)
+
+
+        logger.info(f"[FaceSwap] Incoming request → target_category_id={target_category_id}, new_category_id={new_category_id}, user_id={user_id}")
+
+        if target_category_id and new_category_id:
+            raise HTTPException(400, "Provide only one of new_category_id or target_category_id.")
+
+        if not target_category_id and not new_category_id:
+            raise HTTPException(400, "Either new_category_id or target_category_id is required.")
+
+        # ------------------------------------------------------------------
+        # READ SOURCE IMAGE
+        # ------------------------------------------------------------------
+        src_bytes = await source.read()
+        src_key = f"faceswap/source/{uuid.uuid4().hex}_{source.filename}"
+        upload_to_spaces(src_bytes, src_key, content_type=source.content_type)
+
+        # ------------------------------------------------------------------
+        # CASE 1 : new_category_id → MongoDB lookup
+        # ------------------------------------------------------------------
+        if new_category_id:
+
+            # doc = await subcategories_col.find_one({
+            #     "asset_images._id": ObjectId(new_category_id)
+            # })
+            doc = await subcategories_collection.find_one({
+                "asset_images._id": ObjectId(new_category_id)
+            })
+
+
+            if not doc:
+                raise HTTPException(404, "Asset image not found in database")
+
+            # extract correct asset
+            asset = next(
+                (img for img in doc["asset_images"] if str(img["_id"]) == new_category_id),
+                None
+            )
+
+            if not asset:
+                raise HTTPException(404, "Asset image URL not found")
+
+            # correct URL
+            target_url = asset["url"]
+
+            # correct categoryId (ObjectId)
+            #category_oid = doc["categoryId"]   # <-- DO NOT CONVERT TO STRING
+            subcategory_oid = doc["_id"]
+
+#         # ------------------------------------------------------------------
+#         # CASE 2 : target_category_id → DigitalOcean path (unchanged logic)
+#         # ------------------------------------------------------------------
+        if target_category_id:
+            client = get_spaces_client()
+            base_prefix = "faceswap/target/"
+            resp = client.list_objects_v2(
+                Bucket=DO_SPACES_BUCKET, Prefix=base_prefix, Delimiter="/"
+            )
+            
+            # Extract categories from the CommonPrefixes
+            categories = [p["Prefix"].split("/")[2] for p in resp.get("CommonPrefixes", [])]
+
+            target_url = None
+            
+            # --- FIX STARTS HERE ---
+            for category in categories:
+                original_prefix = f"faceswap/target/{category}/original/"
+                thumb_prefix = f"faceswap/target/{category}/thumb/" # Keep for file list check (optional but safe)
+                
+                # List objects in original/
+                original_objects = client.list_objects_v2(
+                    Bucket=DO_SPACES_BUCKET, Prefix=original_prefix
+                ).get("Contents", [])
+
+                # List objects in thumb/ (optional: for the old code's extra check)
+                thumb_objects = client.list_objects_v2(
+                    Bucket=DO_SPACES_BUCKET, Prefix=thumb_prefix
+                ).get("Contents", [])
+                
+                # Extract only the filenames and filter for .png
+                original_filenames = sorted([
+                    obj["Key"].split("/")[-1] for obj in original_objects 
+                    if obj["Key"].split("/")[-1].endswith(".png")
+                ])
+                thumb_filenames = [
+                    obj["Key"].split("/")[-1] for obj in thumb_objects
+                ]
+
+                # Replicate the old indexing logic based on sorted filenames
+                for idx, filename in enumerate(original_filenames, start=1):
+                    cid = f"{category.lower()}image_{idx}"
+
+                    # Optional: Replicate the thumb file check for 100% parity
+                    # if filename in thumb_filenames and cid == target_category_id:
+                    # Simpler check just on the ID, assuming thumb files are present
+                    if cid == target_category_id:
+                        # Construct the final target URL using the full prefix and the filename
+                        target_url = f"{DO_SPACES_ENDPOINT}/{DO_SPACES_BUCKET}/{original_prefix}{filename}"
+                        break
+                
+                if target_url:
+                    break
+            # --- FIX ENDS HERE ---
+
+            if not target_url:
+                raise HTTPException(404, "Target categoryId not found")
+#         # ------------------------------------------------------------------
+#         # DOWNLOAD TARGET IMAGE
+#         # ------------------------------------------------------------------
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            response = await client.get(target_url)
+            response.raise_for_status()
+            tgt_bytes = response.content
+
+        src_bgr = cv2.imdecode(np.frombuffer(src_bytes, np.uint8), cv2.IMREAD_COLOR)
+        tgt_bgr = cv2.imdecode(np.frombuffer(tgt_bytes, np.uint8), cv2.IMREAD_COLOR)
+
+        if src_bgr is None or tgt_bgr is None:
+            raise HTTPException(400, "Invalid image data")
+
+        src_rgb = cv2.cvtColor(src_bgr, cv2.COLOR_BGR2RGB)
+        tgt_rgb = cv2.cvtColor(tgt_bgr, cv2.COLOR_BGR2RGB)
+
+        # ------------------------------------------------------------------
+        # READ OPTIONAL IMAGE2
+        # ------------------------------------------------------------------
+        img2_rgb = None
+        if image2:
+            img2_bytes = await image2.read()
+            img2_bgr = cv2.imdecode(np.frombuffer(img2_bytes, np.uint8), cv2.IMREAD_COLOR)
+            if img2_bgr is not None:
+                img2_rgb = cv2.cvtColor(img2_bgr, cv2.COLOR_BGR2RGB)
+
+        # ------------------------------------------------------------------
+        # FACE SWAP EXECUTION (run in thread to not block event loop)
+        # ------------------------------------------------------------------
+        if img2_rgb is not None:
+            def _couple_swap():
+                pipeline_start = time.time()
+                src_images = [src_rgb, img2_rgb]
+
+                all_src_faces = []
+                t0 = time.time()
+                for img in src_images:
+                    faces = face_analysis_app.get(cv2.cvtColor(img, cv2.COLOR_RGB2BGR))
+                    all_src_faces.extend(faces)
+                
+                tgt_faces = face_analysis_app.get(cv2.cvtColor(tgt_rgb, cv2.COLOR_RGB2BGR))
+                logger.info(f"[Pipeline] Couple face detection: {time.time()-t0:.2f}s")
+
+                if not all_src_faces:
+                    raise ValueError("No faces detected in source images")
+                if not tgt_faces:
+                    raise ValueError("No faces detected in target image")
+
+                def face_sort_key(face):
+                    x1, y1, x2, y2 = face.bbox
+                    area = (x2 - x1) * (y2 - y1)
+                    cx = (x1 + x2) / 2
+                    return (-area, cx)
+
+                src_male = sorted([f for f in all_src_faces if f.gender == 1], key=face_sort_key)
+                src_female = sorted([f for f in all_src_faces if f.gender == 0], key=face_sort_key)
+                tgt_male = sorted([f for f in tgt_faces if f.gender == 1], key=face_sort_key)
+                tgt_female = sorted([f for f in tgt_faces if f.gender == 0], key=face_sort_key)
+
+                pairs = []
+                for s, t in zip(src_male, tgt_male):
+                    pairs.append((s, t))
+                for s, t in zip(src_female, tgt_female):
+                    pairs.append((s, t))
+
+                if not pairs:
+                    src_all = sorted(all_src_faces, key=face_sort_key)
+                    tgt_all = sorted(tgt_faces, key=face_sort_key)
+                    pairs = list(zip(src_all, tgt_all))
+
+                t0 = time.time()
+                with swap_lock:
+                    result_img = cv2.cvtColor(tgt_rgb, cv2.COLOR_RGB2BGR)
+                    for src_face, _ in pairs:
+                        current_faces = sorted(face_analysis_app.get(result_img), key=face_sort_key)
+                        candidates = [f for f in current_faces if f.gender == src_face.gender] or current_faces
+                        target_face = candidates[0]
+                        result_img = swapper.get(result_img, target_face, src_face, paste_back=True)
+                logger.info(f"[Pipeline] Couple face swap: {time.time()-t0:.2f}s")
+
+                result_rgb_out = cv2.cvtColor(result_img, cv2.COLOR_BGR2RGB)
+
+                t0 = time.time()
+                enhanced_rgb = mandatory_enhancement(result_rgb_out)
+                logger.info(f"[Pipeline] Couple enhancement: {time.time()-t0:.2f}s")
+                
+                enhanced_bgr = cv2.cvtColor(enhanced_rgb, cv2.COLOR_RGB2BGR)
+
+                temp_dir = tempfile.mkdtemp(prefix="faceswap_")
+                final_path = os.path.join(temp_dir, "result.png")
+                cv2.imwrite(final_path, enhanced_bgr)
+
+                with open(final_path, "rb") as f:
+                    result_bytes = f.read()
+                
+                logger.info(f"[Pipeline] TOTAL couple swap: {time.time()-pipeline_start:.2f}s")
+                return result_bytes
+
+            try:
+                result_bytes = await asyncio.to_thread(_couple_swap)
+            except ValueError as ve:
+                raise HTTPException(400, str(ve))
+
+        else:
+            # ----- SINGLE SOURCE SWAP (run in thread) -----
+            def _single_swap():
+                return face_swap_and_enhance(src_rgb, tgt_rgb)
+            
+            final_img, final_path, err = await asyncio.to_thread(_single_swap)
+            
+            if err:
+                raise HTTPException(500, err)
+
+            with open(final_path, "rb") as f:
+                result_bytes = f.read()
+
+        result_key = f"faceswap/result/{uuid.uuid4().hex}_enhanced.png"
+        result_url = upload_to_spaces(result_bytes, result_key)
+        # -------------------------------------------------
+        # COMPRESS IMAGE (2–3 MB target)
+        # -------------------------------------------------
+        compressed_bytes = compress_image(
+            image_bytes=result_bytes,
+            max_size=(1280, 1280),
+            quality=72
+        )
+        
+        compressed_key = f"faceswap/result/{uuid.uuid4().hex}_enhanced_compressed.jpg"
+        compressed_url = upload_to_spaces(
+            compressed_bytes,
+            compressed_key,
+            content_type="image/jpeg"
+        )
+        end_time = datetime.utcnow()
+        response_time_ms = (end_time - start_time).total_seconds() * 1000
+
+        # if database is not None:
+        #     log_entry = {
+        #         "endpoint": "/face-swap",
+        #         "status": "success",
+        #         "response_time_ms": response_time_ms,
+        #         "timestamp": end_time
+        #     }
+        #     if appname:
+        #         log_entry["appname"] = appname
+        #     await database.api_logs.insert_one(log_entry)
+        if logs_collection is not None:
+            log_entry = {
+                "endpoint": "/face-swap",
+                "status": "success",
+                "response_time_ms": float(response_time_ms),
+                "timestamp": end_time,
+                "appname": appname if appname else None,
+                "error": None
+            }
+        
+            await logs_collection.insert_one(log_entry)
+
+
+        return {
+            "result_key": result_key,
+            "result_url": result_url,
+            "Compressed_Image_URL": compressed_url
+        }
+        
+    except Exception as e:
+        end_time = datetime.utcnow()
+        response_time_ms = (end_time - start_time).total_seconds() * 1000
+
+        # if database is not None:
+        #     log_entry = {
+        #         "endpoint": "/face-swap",
+        #         "status": "fail",
+        #         "response_time_ms": response_time_ms,
+        #         "timestamp": end_time,
+        #         "error": str(e)
+        #     }
+        #     if appname:
+        #         log_entry["appname"] = appname
+        #     await database.api_logs.insert_one(log_entry)
+        if logs_collection is not None:
+            log_entry = {
+                "endpoint": "/face-swap",
+                "status": "fail",
+                "response_time_ms": float(response_time_ms),
+                "timestamp": end_time,
+                "appname": appname if appname else None,
+                "error": str(e)
+            }
+        
+            await logs_collection.insert_one(log_entry)
+
+        raise HTTPException(500, f"Face swap failed: {str(e)}")
+
+@fastapi_app.get("/preview/{result_key:path}")
+async def preview_result(result_key: str):
+    try:
+        img_bytes = download_from_spaces(result_key)
+    except Exception:
+        raise HTTPException(status_code=404, detail="Result not found")
+    return Response(
+        content=img_bytes,
+        media_type="image/png",
+        headers={"Content-Disposition": "inline; filename=result.png"}
+    )
+    
+@fastapi_app.post("/multi-face-swap", dependencies=[Depends(verify_token)])
+async def multi_face_swap_api(
+    source_image: UploadFile = File(...),
+    target_image: UploadFile = File(...)
+):
+    start_time = datetime.utcnow()
+
+    try:
+        # -----------------------------
+        # Read images
+        # -----------------------------
+        src_bytes = await source_image.read()
+        tgt_bytes = await target_image.read()
+
+        src_bgr = cv2.imdecode(np.frombuffer(src_bytes, np.uint8), cv2.IMREAD_COLOR)
+        tgt_bgr = cv2.imdecode(np.frombuffer(tgt_bytes, np.uint8), cv2.IMREAD_COLOR)
+
+        if src_bgr is None or tgt_bgr is None:
+            raise HTTPException(400, "Invalid image data")
+
+        src_rgb = cv2.cvtColor(src_bgr, cv2.COLOR_BGR2RGB)
+        tgt_rgb = cv2.cvtColor(tgt_bgr, cv2.COLOR_BGR2RGB)
+
+        # -----------------------------
+        # Multi-face swap (run in thread to not block event loop)
+        # -----------------------------
+        def _multi_swap_and_enhance():
+            swapped_rgb = multi_face_swap(src_rgb, tgt_rgb)
+            return mandatory_enhancement(swapped_rgb)
+        
+        final_rgb = await asyncio.to_thread(_multi_swap_and_enhance)
+
+        final_bgr = cv2.cvtColor(final_rgb, cv2.COLOR_RGB2BGR)
+
+        # -----------------------------
+        # Save temp result
+        # -----------------------------
+        temp_dir = tempfile.mkdtemp(prefix="multi_faceswap_")
+        result_path = os.path.join(temp_dir, "result.png")
+        cv2.imwrite(result_path, final_bgr)
+
+        with open(result_path, "rb") as f:
+            result_bytes = f.read()
+
+        # -----------------------------
+        # Upload
+        # -----------------------------
+        result_key = f"faceswap/multi/{uuid.uuid4().hex}.png"
+        result_url = upload_to_spaces(
+            result_bytes,
+            result_key,
+            content_type="image/png"
+        )
+
+        return {
+            "result_key": result_key,
+            "result_url": result_url
+        }
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))
+
+
+@fastapi_app.post("/face-swap-couple", dependencies=[Depends(verify_token)])
+async def face_swap_couple_api(
+    image1: UploadFile = File(...),
+    image2: Optional[UploadFile] = File(None),
+    target_category_id: str = Form(None),
+    new_category_id: str = Form(None),
+    user_id: Optional[str] = Form(None),
+    appname: Optional[str] = Form(None),
+    credentials: HTTPAuthorizationCredentials = Security(security)
+):
+    """
+    Production-ready face swap endpoint supporting:
+    - Multiple source images (image1 + optional image2)
+    - Gender-based pairing
+    - Merged faces from multiple sources
+    - Mandatory CodeFormer enhancement
+    """
+    start_time = datetime.utcnow()
+    
+    try:
+        # -----------------------------
+        # Validate input
+        # -----------------------------
+        if target_category_id == "":
+            target_category_id = None
+        if new_category_id == "":
+            new_category_id = None
+        if user_id == "":
+            user_id = None
+
+        subcategories_collection = get_app_db_collections(appname)
+        logger.info(f"[FaceSwapCouple] appname={appname}, subcategories_collection={subcategories_collection.full_name if subcategories_collection is not None else 'None'}")
+
+        if target_category_id and new_category_id:
+            raise HTTPException(400, "Provide only one of new_category_id or target_category_id.")
+        if not target_category_id and not new_category_id:
+            raise HTTPException(400, "Either new_category_id or target_category_id is required.")
+
+        logger.info(f"[FaceSwap] Incoming request → target_category_id={target_category_id}, new_category_id={new_category_id}, user_id={user_id}")
+
+        # -----------------------------
+        # Read source images
+        # -----------------------------
+        src_images = []
+        img1_bytes = await image1.read()
+        src1 = cv2.imdecode(np.frombuffer(img1_bytes, np.uint8), cv2.IMREAD_COLOR)
+        if src1 is None:
+            raise HTTPException(400, "Invalid image1 data")
+        src_images.append(cv2.cvtColor(src1, cv2.COLOR_BGR2RGB))
+
+        if image2:
+            img2_bytes = await image2.read()
+            src2 = cv2.imdecode(np.frombuffer(img2_bytes, np.uint8), cv2.IMREAD_COLOR)
+            if src2 is not None:
+                src_images.append(cv2.cvtColor(src2, cv2.COLOR_BGR2RGB))
+
+        # -----------------------------
+        # Resolve target image
+        # -----------------------------
+        target_url = None
+        if new_category_id:
+            # doc = await subcategories_col.find_one({
+            #     "asset_images._id": ObjectId(new_category_id)
+            # })
+            doc = await subcategories_collection.find_one({
+                "asset_images._id": ObjectId(new_category_id)
+            })
+
+            if not doc:
+                raise HTTPException(404, "Asset image not found in database")
+
+            asset = next(
+                (img for img in doc["asset_images"] if str(img["_id"]) == new_category_id),
+                None
+            )
+
+            if not asset:
+                raise HTTPException(404, "Asset image URL not found")
+
+            target_url = asset["url"]
+            subcategory_oid = doc["_id"]
+
+        if target_category_id:
+            client = get_spaces_client()
+            base_prefix = "faceswap/target/"
+            resp = client.list_objects_v2(
+                Bucket=DO_SPACES_BUCKET, Prefix=base_prefix, Delimiter="/"
+            )
+
+            categories = [p["Prefix"].split("/")[2] for p in resp.get("CommonPrefixes", [])]
+
+            for category in categories:
+                original_prefix = f"faceswap/target/{category}/original/"
+                thumb_prefix = f"faceswap/target/{category}/thumb/"
+
+                original_objects = client.list_objects_v2(
+                    Bucket=DO_SPACES_BUCKET, Prefix=original_prefix
+                ).get("Contents", [])
+
+                thumb_objects = client.list_objects_v2(
+                    Bucket=DO_SPACES_BUCKET, Prefix=thumb_prefix
+                ).get("Contents", [])
+
+                original_filenames = sorted([
+                    obj["Key"].split("/")[-1] for obj in original_objects
+                    if obj["Key"].split("/")[-1].endswith(".png")
+                ])
+
+                for idx, filename in enumerate(original_filenames, start=1):
+                    cid = f"{category.lower()}image_{idx}"
+                    if cid == target_category_id:
+                        target_url = f"{DO_SPACES_ENDPOINT}/{DO_SPACES_BUCKET}/{original_prefix}{filename}"
+                        break
+
+                if target_url:
+                    break
+
+            if not target_url:
+                raise HTTPException(404, "Target categoryId not found")
+
+        async with httpx.AsyncClient(timeout=30.0) as client:
+            response = await client.get(target_url)
+            response.raise_for_status()
+            tgt_bytes = response.content
+
+        tgt_bgr = cv2.imdecode(np.frombuffer(tgt_bytes, np.uint8), cv2.IMREAD_COLOR)
+        if tgt_bgr is None:
+            raise HTTPException(400, "Invalid target image data")
+
+        # -----------------------------
+        # Couple face swap + enhance (run in thread)
+        # -----------------------------
+        def _couple_face_swap_and_enhance():
+            pipeline_start = time.time()
+            
+            all_src_faces = []
+            t0 = time.time()
+            for img in src_images:
+                faces = face_analysis_app.get(cv2.cvtColor(img, cv2.COLOR_RGB2BGR))
+                all_src_faces.extend(faces)
+
+            tgt_faces = face_analysis_app.get(tgt_bgr)
+            logger.info(f"[Pipeline] Couple-ep face detection: {time.time()-t0:.2f}s")
+
+            if not all_src_faces:
+                raise ValueError("No faces detected in source images")
+            if not tgt_faces:
+                raise ValueError("No faces detected in target image")
+
+            def face_sort_key(face):
+                x1, y1, x2, y2 = face.bbox
+                area = (x2 - x1) * (y2 - y1)
+                cx = (x1 + x2) / 2
+                return (-area, cx)
+
+            src_male = sorted([f for f in all_src_faces if f.gender == 1], key=face_sort_key)
+            src_female = sorted([f for f in all_src_faces if f.gender == 0], key=face_sort_key)
+            tgt_male = sorted([f for f in tgt_faces if f.gender == 1], key=face_sort_key)
+            tgt_female = sorted([f for f in tgt_faces if f.gender == 0], key=face_sort_key)
+
+            pairs = []
+            for s, t in zip(src_male, tgt_male):
+                pairs.append((s, t))
+            for s, t in zip(src_female, tgt_female):
+                pairs.append((s, t))
+
+            if not pairs:
+                src_all = sorted(all_src_faces, key=face_sort_key)
+                tgt_all = sorted(tgt_faces, key=face_sort_key)
+                pairs = list(zip(src_all, tgt_all))
+
+            t0 = time.time()
+            with swap_lock:
+                result_img = tgt_bgr.copy()
+                for src_face, _ in pairs:
+                    current_faces = sorted(face_analysis_app.get(result_img), key=face_sort_key)
+                    candidates = [f for f in current_faces if f.gender == src_face.gender] or current_faces
+                    target_face = candidates[0]
+                    result_img = swapper.get(result_img, target_face, src_face, paste_back=True)
+            logger.info(f"[Pipeline] Couple-ep face swap: {time.time()-t0:.2f}s")
+
+            result_rgb = cv2.cvtColor(result_img, cv2.COLOR_BGR2RGB)
+
+            t0 = time.time()
+            enhanced_rgb = mandatory_enhancement(result_rgb)
+            logger.info(f"[Pipeline] Couple-ep enhancement: {time.time()-t0:.2f}s")
+            
+            enhanced_bgr = cv2.cvtColor(enhanced_rgb, cv2.COLOR_RGB2BGR)
+
+            temp_dir = tempfile.mkdtemp(prefix="faceswap_")
+            final_path = os.path.join(temp_dir, "result.png")
+            cv2.imwrite(final_path, enhanced_bgr)
+
+            with open(final_path, "rb") as f:
+                result_bytes = f.read()
+            
+            logger.info(f"[Pipeline] TOTAL couple-ep swap: {time.time()-pipeline_start:.2f}s")
+            return result_bytes
+
+        try:
+            result_bytes = await asyncio.to_thread(_couple_face_swap_and_enhance)
+        except ValueError as ve:
+            raise HTTPException(400, str(ve))
+
+        result_key = f"faceswap/result/{uuid.uuid4().hex}_enhanced.png"
+        result_url = upload_to_spaces(result_bytes, result_key)
+
+        compressed_bytes = compress_image(result_bytes, max_size=(1280, 1280), quality=72)
+        compressed_key = f"faceswap/result/{uuid.uuid4().hex}_enhanced_compressed.jpg"
+        compressed_url = upload_to_spaces(compressed_bytes, compressed_key, content_type="image/jpeg")
+
+        # -----------------------------
+        # Log API usage
+        # -----------------------------
+        end_time = datetime.utcnow()
+        response_time_ms = (end_time - start_time).total_seconds() * 1000
+        
+        if logs_collection is not None:
+            log_entry = {
+                "endpoint": "/face-swap-couple",
+                "status": "success",
+                "response_time_ms": float(response_time_ms),
+                "timestamp": end_time,
+                "appname": appname if appname else None,
+                "error": None
+            }
+        
+            await logs_collection.insert_one(log_entry)
+
+        return {
+            "result_key": result_key,
+            "result_url": result_url,
+            "compressed_url": compressed_url
+        }
+
+    except Exception as e:
+        end_time = datetime.utcnow()
+        response_time_ms = (end_time - start_time).total_seconds() * 1000
+      
+        if logs_collection is not None:
+            log_entry = {
+                "endpoint": "/face-swap-couple",
+                "status": "fail",
+                "response_time_ms": float(response_time_ms),
+                "timestamp": end_time,
+                "appname": appname if appname else None,
+                "error": str(e)
+            }
+        
+            await logs_collection.insert_one(log_entry)
+        raise HTTPException(500, f"Face swap failed: {str(e)}")
+
+
+if __name__ == "__main__":
+    uvicorn.run(fastapi_app, host="0.0.0.0", port=7860)