import io import os import uuid import logging from typing import Optional from fastapi import FastAPI, UploadFile, File, HTTPException, Depends, Header from fastapi.responses import FileResponse, JSONResponse from pydantic import BaseModel import torch import asyncio import time import numpy as np from PIL import Image # Lazy import performed in get_model() to avoid import-time failures on Space LOGGER = logging.getLogger("hair_server") logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(name)s - %(message)s") EXPECTED_BEARER = "logicgo@123" # Optional Mongo persistence from pymongo import MongoClient MONGO_URI = os.environ.get("MONGO_URI", "") mongo_client = MongoClient(MONGO_URI) if MONGO_URI else None mongo_db = mongo_client.get_database("HairSwapDB") if mongo_client else None uploads_col = mongo_db.get_collection("uploads") if mongo_db else None results_col = mongo_db.get_collection("results") if mongo_db else None def verify_bearer(authorization: Optional[str] = Header(None)): if not authorization: raise HTTPException(status_code=401, detail="Missing Authorization header") try: scheme, token = authorization.split(" ", 1) except ValueError: raise HTTPException(status_code=401, detail="Invalid Authorization header format") if scheme.lower() != "bearer": raise HTTPException(status_code=401, detail="Invalid auth scheme") if token != EXPECTED_BEARER: raise HTTPException(status_code=401, detail="Invalid token") return True app = FastAPI(title="Hair Swap API", version="1.0.0") @app.get("/health") def health(): return {"status": "healthy"} @app.get("/") def root(): return {"status": "ok"} @app.get("/health/gpu") def health_gpu(): try: import torch as _t return { "cuda_available": bool(_t.cuda.is_available()), "cuda_device_count": int(_t.cuda.device_count()) if _t.cuda.is_available() else 0, "cuda_device_name": _t.cuda.get_device_name(0) if _t.cuda.is_available() else None, "torch_version": getattr(_t, "__version__", None), "torch_cuda_version": getattr(_t.version, "cuda", None), } except Exception as e: return {"error": str(e)} class HairSwapRequest(BaseModel): source_id: str reference_id: str converter_scale: float = 1.0 scale: float = 1.0 guidance_scale: float = 1.5 controlnet_conditioning_scale: float = 1.0 # Initialize model lazily on first request _model = None # type: ignore[assignment] def get_model(): global _model if _model is None: try: LOGGER.info("Loading StableHair model ...") # Enforce GPU-only execution on Spaces with L4/other CUDA GPUs if not torch.cuda.is_available(): raise RuntimeError("CUDA GPU not available. Install CUDA-enabled PyTorch and run on a GPU Space.") device = "cuda" dtype = torch.float16 # Speed knobs for NVIDIA GPUs try: torch.backends.cuda.matmul.allow_tf32 = True # type: ignore[attr-defined] torch.backends.cudnn.allow_tf32 = True # type: ignore[attr-defined] torch.backends.cudnn.benchmark = True # type: ignore[attr-defined] except Exception: pass LOGGER.info(f"Using device: {device}, dtype: {dtype}") # Ensure HF token env var is where downstream libs expect it if os.environ.get("HUGGINGFACEHUB_API_TOKEN") and not os.environ.get("HUGGINGFACE_HUB_TOKEN"): os.environ["HUGGINGFACE_HUB_TOKEN"] = os.environ["HUGGINGFACEHUB_API_TOKEN"] # Backward-compat shim: some diffusers versions import a helper only present in newer hub versions. try: import huggingface_hub as _hfh # type: ignore if not hasattr(_hfh, "split_torch_state_dict_into_shards"): def _split_torch_state_dict_into_shards(state_dict, max_shard_size="10GB"): # Minimal shim: return a single shard mapping expected by callers return {"pytorch_model.bin": state_dict} _hfh.split_torch_state_dict_into_shards = _split_torch_state_dict_into_shards # type: ignore[attr-defined] except Exception: pass # Import here to defer importing diffusers/transformers until needed from infer_full import StableHair # noqa: WPS433 _model = StableHair(config="./configs/hair_transfer.yaml", device=device, weight_dtype=dtype) LOGGER.info("Model loaded successfully") except Exception as e: LOGGER.error(f"Failed to load model: {str(e)}") raise Exception(f"Model loading failed: {str(e)}") return _model # Use a writable location on Hugging Face Spaces BASE_DATA_DIR = os.environ.get("DATA_DIR", "/data") UPLOAD_DIR = os.path.join(BASE_DATA_DIR, "uploads") RESULTS_DIR = os.path.join(BASE_DATA_DIR, "results") LOGS_DIR = os.path.join(BASE_DATA_DIR, "logs") os.makedirs(UPLOAD_DIR, exist_ok=True) os.makedirs(RESULTS_DIR, exist_ok=True) os.makedirs(LOGS_DIR, exist_ok=True) @app.post("/upload") async def upload_image(image: UploadFile = File(...), _=Depends(verify_bearer)): if not image.filename: raise HTTPException(status_code=400, detail="No file name provided") contents = await image.read() try: Image.open(io.BytesIO(contents)).convert("RGB") except Exception: raise HTTPException(status_code=400, detail="Invalid image file") image_id = str(uuid.uuid4()) ext = os.path.splitext(image.filename)[1] or ".png" path = os.path.join(UPLOAD_DIR, image_id + ext) with open(path, "wb") as f: f.write(contents) # Save metadata to Mongo if uploads_col: try: uploads_col.insert_one({"_id": image_id, "filename": os.path.basename(path), "path": path}) except Exception: pass return {"id": image_id, "filename": os.path.basename(path)} @app.post("/get-hairswap") def get_hairswap(req: HairSwapRequest, _=Depends(verify_bearer)): try: # Resolve file paths def find_file(image_id: str) -> str: for name in os.listdir(UPLOAD_DIR): if name.startswith(image_id): return os.path.join(UPLOAD_DIR, name) raise HTTPException(status_code=404, detail=f"Image id not found: {image_id}") source_path = find_file(req.source_id) reference_path = find_file(req.reference_id) LOGGER.info(f"Found source: {source_path}, reference: {reference_path}") # Load model with error handling try: model = get_model() LOGGER.info("Model loaded successfully") except Exception as e: LOGGER.error(f"Model loading failed: {str(e)}") raise HTTPException(status_code=500, detail=f"Model loading failed: {str(e)}") # Perform hair transfer with error handling try: LOGGER.info("Starting hair transfer...") # Log current schedulers for visibility try: sched_main = type(model.pipeline.scheduler).__name__ if hasattr(model, "pipeline") else None sched_bald = type(model.remove_hair_pipeline.scheduler).__name__ if hasattr(model, "remove_hair_pipeline") else None LOGGER.info(f"Schedulers -> main: {sched_main}, remove_hair: {sched_bald}") except Exception: pass id_np, out_np, bald_np, ref_np = model.Hair_Transfer( source_image=source_path, reference_image=reference_path, random_seed=-1, step=20, guidance_scale=req.guidance_scale, scale=req.scale, controlnet_conditioning_scale=req.controlnet_conditioning_scale, size=448, ) LOGGER.info("Hair transfer completed successfully") except Exception as e: import traceback tb = traceback.format_exc() LOGGER.error(f"Hair transfer failed: {str(e)} | device={model.device if hasattr(model, 'device') else 'n/a'} cuda_available={torch.cuda.is_available()}\n{tb}") raise HTTPException(status_code=500, detail=f"Hair transfer failed: {str(e)}") # Save result try: result_id = str(uuid.uuid4()) out_img = Image.fromarray((out_np * 255.).astype(np.uint8)) filename = f"{result_id}.png" out_path = os.path.join(RESULTS_DIR, filename) out_img.save(out_path) LOGGER.info(f"Result saved: {out_path}") if results_col: try: results_col.insert_one({ "_id": result_id, "filename": filename, "path": out_path, "source_id": req.source_id, "reference_id": req.reference_id, }) except Exception as e: LOGGER.warning(f"MongoDB save failed: {str(e)}") return {"result": filename} except Exception as e: LOGGER.error(f"Result saving failed: {str(e)}") raise HTTPException(status_code=500, detail=f"Result saving failed: {str(e)}") except HTTPException: raise except Exception as e: LOGGER.error(f"Unexpected error in get_hairswap: {str(e)}") raise HTTPException(status_code=500, detail=f"Unexpected error: {str(e)}") @app.get("/download/{filename}") def download(filename: str, _=Depends(verify_bearer)): path = os.path.join(RESULTS_DIR, filename) if not os.path.exists(path): raise HTTPException(status_code=404, detail="File not found") return FileResponse(path, media_type="image/png", filename=filename) @app.get("/logs") def logs(_=Depends(verify_bearer)): if uploads_col and results_col: uploads = list(uploads_col.find({}, {"_id": 1, "filename": 1}).limit(20)) results = list(results_col.find({}, {"_id": 1, "filename": 1, "source_id": 1, "reference_id": 1}).limit(20)) return JSONResponse({"uploads": uploads, "results": results}) return JSONResponse({"logs": ["service running"], "db": "not_configured"}) # -------------------- Async job API -------------------- MAX_CONCURRENCY = int(os.environ.get("MAX_CONCURRENCY", "1")) _sem = asyncio.Semaphore(MAX_CONCURRENCY) _jobs = {} @app.post("/get-hairswap-async") async def get_hairswap_async(req: HairSwapRequest, _=Depends(verify_bearer)): job_id = str(uuid.uuid4()) _jobs[job_id] = {"status": "queued", "result": None, "error": None, "started_at": None, "ended_at": None} async def _run_job(): _jobs[job_id]["status"] = "running" _jobs[job_id]["started_at"] = time.time() try: async with _sem: # reuse the same core flow as sync endpoint def find_file(image_id: str) -> str: for name in os.listdir(UPLOAD_DIR): if name.startswith(image_id): return os.path.join(UPLOAD_DIR, name) raise HTTPException(status_code=404, detail=f"Image id not found: {image_id}") source_path = find_file(req.source_id) reference_path = find_file(req.reference_id) LOGGER.info(f"[job {job_id}] Found source: {source_path}, reference: {reference_path}") model = get_model() LOGGER.info(f"[job {job_id}] Model loaded successfully") LOGGER.info(f"[job {job_id}] Starting hair transfer...") try: try: sched_main = type(model.pipeline.scheduler).__name__ if hasattr(model, "pipeline") else None sched_bald = type(model.remove_hair_pipeline.scheduler).__name__ if hasattr(model, "remove_hair_pipeline") else None LOGGER.info(f"[job {job_id}] Schedulers -> main: {sched_main}, remove_hair: {sched_bald}") except Exception: pass id_np, out_np, bald_np, ref_np = model.Hair_Transfer( source_image=source_path, reference_image=reference_path, random_seed=-1, step=20, guidance_scale=req.guidance_scale, scale=req.scale, controlnet_conditioning_scale=req.controlnet_conditioning_scale, size=448, ) result_id = str(uuid.uuid4()) out_img = Image.fromarray((out_np * 255.).astype(np.uint8)) filename = f"{result_id}.png" out_path = os.path.join(RESULTS_DIR, filename) out_img.save(out_path) _jobs[job_id]["result"] = {"filename": filename} _jobs[job_id]["status"] = "completed" LOGGER.info(f"[job {job_id}] Completed -> {out_path}") except Exception as e: LOGGER.error(f"[job {job_id}] Hair transfer failed: {str(e)}") _jobs[job_id]["error"] = str(e) _jobs[job_id]["status"] = "failed" except Exception as e: _jobs[job_id]["error"] = str(e) _jobs[job_id]["status"] = "failed" finally: _jobs[job_id]["ended_at"] = time.time() asyncio.create_task(_run_job()) return {"job_id": job_id, "status": "queued"} @app.get("/job/{job_id}") def job_status(job_id: str, _=Depends(verify_bearer)): data = _jobs.get(job_id) if not data: raise HTTPException(status_code=404, detail="job not found") return data