api.py

from __future__ import annotations
from fastapi import FastAPI, UploadFile, File, Form, BackgroundTasks, HTTPException
from fastapi.responses import JSONResponse, FileResponse
from fastapi.middleware.cors import CORSMiddleware
from pathlib import Path
import shutil
import uvicorn
import json
import uuid
from datetime import datetime
from typing import Dict
from enum import Enum
import os

from video_processing import process_video_pipeline
from casting_loader import ensure_chroma, build_faces_index, build_voices_index
from narration_system import NarrationSystem
from llm_router import load_yaml, LLMRouter
from character_detection import detect_characters_from_video

app = FastAPI(title="Veureu Engine API", version="0.2.0")
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

ROOT = Path("/tmp/veureu")
ROOT.mkdir(parents=True, exist_ok=True)
TEMP_ROOT = Path("/tmp/temp")
TEMP_ROOT.mkdir(parents=True, exist_ok=True)
VIDEOS_ROOT = Path("/tmp/data/videos")
VIDEOS_ROOT.mkdir(parents=True, exist_ok=True)

# Sistema de jobs asíncronos
class JobStatus(str, Enum):
    QUEUED = "queued"
    PROCESSING = "processing"
    DONE = "done"
    FAILED = "failed"

jobs: Dict[str, dict] = {}

@app.get("/")
def root():
    return {"ok": True, "service": "veureu-engine"}

@app.post("/process_video")
async def process_video(
    video_file: UploadFile = File(...),
    config_path: str = Form("config.yaml"),
    out_root: str = Form("results"),
    db_dir: str = Form("chroma_db"),
):
    tmp_video = ROOT / video_file.filename
    with tmp_video.open("wb") as f:
        shutil.copyfileobj(video_file.file, f)
    result = process_video_pipeline(str(tmp_video), config_path=config_path, out_root=out_root, db_dir=db_dir)
    return JSONResponse(result)

@app.post("/create_initial_casting")
async def create_initial_casting(
    background_tasks: BackgroundTasks,
    video: UploadFile = File(...),
    epsilon: float = Form(...),
    min_cluster_size: int = Form(...),
):
    """
    Crea un job para procesar el vídeo de forma asíncrona.
    Devuelve un job_id inmediatamente.
    """
    # Guardar vídeo en carpeta de datos
    video_name = Path(video.filename).stem
    dst_video = VIDEOS_ROOT / f"{video_name}.mp4"
    with dst_video.open("wb") as f:
        shutil.copyfileobj(video.file, f)

    # Crear job_id único
    job_id = str(uuid.uuid4())
    
    # Inicializar el job
    jobs[job_id] = {
        "id": job_id,
        "status": JobStatus.QUEUED,
        "video_path": str(dst_video),
        "video_name": video_name,
        "epsilon": float(epsilon),
        "min_cluster_size": int(min_cluster_size),
        "created_at": datetime.now().isoformat(),
        "results": None,
        "error": None
    }
    
    print(f"[{job_id}] Job creado para vídeo: {video_name}")
    
    # Iniciar procesamiento en background
    background_tasks.add_task(process_video_job, job_id)
    
    # Devolver job_id inmediatamente
    return {"job_id": job_id}

@app.get("/jobs/{job_id}/status")
def get_job_status(job_id: str):
    """
    Devuelve el estado actual de un job.
    El UI hace polling de este endpoint cada 5 segundos.
    """
    if job_id not in jobs:
        raise HTTPException(status_code=404, detail="Job not found")
    
    job = jobs[job_id]
    
    response = {
        "status": job["status"]
    }
    
    # Si está completado, incluir resultados
    if job["status"] == JobStatus.DONE:
        response["results"] = job["results"]
    
    # Si falló, incluir error
    elif job["status"] == JobStatus.FAILED:
        response["error"] = job["error"]
    
    return response

@app.get("/files/{video_name}/{char_id}/{filename}")
def serve_character_file(video_name: str, char_id: str, filename: str):
    """
    Sirve archivos estáticos de personajes (imágenes).
    Ejemplo: /files/dif_catala_1/char1/representative.jpg
    """
    file_path = TEMP_ROOT / video_name / char_id / filename
    
    if not file_path.exists():
        raise HTTPException(status_code=404, detail="File not found")
    
    return FileResponse(file_path)

def process_video_job(job_id: str):
    """
    Procesa el vídeo de forma asíncrona.
    Esta función se ejecuta en background.
    """
    try:
        job = jobs[job_id]
        print(f"[{job_id}] Iniciando procesamiento...")
        
        # Cambiar estado a processing
        job["status"] = JobStatus.PROCESSING
        
        video_path = job["video_path"]
        video_name = job["video_name"]
        epsilon = job["epsilon"]
        min_cluster_size = job["min_cluster_size"]
        
        # Crear estructura de carpetas
        base = TEMP_ROOT / video_name
        base.mkdir(parents=True, exist_ok=True)
        
        print(f"[{job_id}] Directorio base: {base}")
        
        # Detección real de personajes usando el código de Ana
        try:
            print(f"[{job_id}] Iniciando detección de personajes...")
            result = detect_characters_from_video(
                video_path=video_path,
                output_base=str(base),
                epsilon=epsilon,
                min_cluster_size=min_cluster_size,
                video_name=video_name
            )
            
            characters = result.get("characters", [])
            analysis_path = result.get("analysis_path", "")
            
            print(f"[{job_id}] Personajes detectados: {len(characters)}")
            for char in characters:
                print(f"[{job_id}]   - {char['name']}: {char['num_faces']} caras")
            
            # Marcar como completado
            job["status"] = JobStatus.DONE
            job["results"] = {
                "characters": characters,
                "num_characters": len(characters),
                "analysis_path": analysis_path,
                "base_dir": str(base)
            }
            
        except Exception as e_detect:
            # Si falla la detección, intentar modo fallback
            print(f"[{job_id}] Error en detección: {e_detect}")
            print(f"[{job_id}] Usando modo fallback (carpetas vacías)")
            
            # Crear carpetas básicas como fallback
            for sub in ("sources", "faces", "voices", "backgrounds"):
                (base / sub).mkdir(parents=True, exist_ok=True)
            
            job["status"] = JobStatus.DONE
            job["results"] = {
                "characters": [],
                "num_characters": 0,
                "temp_dirs": {
                    "sources": str(base / "sources"),
                    "faces": str(base / "faces"),
                    "voices": str(base / "voices"),
                    "backgrounds": str(base / "backgrounds"),
                },
                "warning": f"Detección falló, usando modo fallback: {str(e_detect)}"
            }
        
        print(f"[{job_id}] ✓ Job completado exitosamente")
        
    except Exception as e:
        print(f"[{job_id}] ✗ Error en el procesamiento: {e}")
        jobs[job_id]["status"] = JobStatus.FAILED
        jobs[job_id]["error"] = str(e)

@app.post("/load_casting")
async def load_casting(
    faces_dir: str = Form("identities/faces"),
    voices_dir: str = Form("identities/voices"),
    db_dir: str = Form("chroma_db"),
    drop_collections: bool = Form(False),
):
    client = ensure_chroma(Path(db_dir))
    n_faces = build_faces_index(Path(faces_dir), client, collection_name="index_faces", drop=drop_collections)
    n_voices = build_voices_index(Path(voices_dir), client, collection_name="index_voices", drop=drop_collections)
    return {"ok": True, "faces": n_faces, "voices": n_voices}

@app.post("/refine_narration")
async def refine_narration(
    dialogues_srt: str = Form(...),
    frame_descriptions_json: str = Form("[]"),
    config_path: str = Form("config.yaml"),
):
    cfg = load_yaml(config_path)
    frames = json.loads(frame_descriptions_json)
    model_name = cfg.get("narration", {}).get("model", "salamandra-instruct")
    use_remote = model_name in (cfg.get("models", {}).get("routing", {}).get("use_remote_for", []))

    if use_remote:
        router = LLMRouter(cfg)
        system_msg = (
            "Eres un sistema de audiodescripción que cumple UNE-153010. "
            "Fusiona diálogos del SRT con descripciones concisas en los huecos, evitando redundancias. "
            "Devuelve JSON con {narrative_text, srt_text}."
        )
        prompt = json.dumps({"dialogues_srt": dialogues_srt, "frames": frames, "rules": cfg.get("narration", {})}, ensure_ascii=False)
        try:
            txt = router.instruct(prompt=prompt, system=system_msg, model=model_name)
            out = {}
            try:
                out = json.loads(txt)
            except Exception:
                out = {"narrative_text": txt, "srt_text": ""}
            return {
                "narrative_text": out.get("narrative_text", ""),
                "srt_text": out.get("srt_text", ""),
                "approved": True,
                "critic_feedback": "",
            }
        except Exception:
            ns = NarrationSystem(model_url=None, une_guidelines_path=cfg.get("narration", {}).get("narration_une_guidelines_path", "UNE_153010.txt"))
            res = ns.run(dialogues_srt, frames)
            return {"narrative_text": res.narrative_text, "srt_text": res.srt_text, "approved": res.approved, "critic_feedback": res.critic_feedback}

    ns = NarrationSystem(model_url=None, une_guidelines_path=cfg.get("narration", {}).get("une_guidelines_path", "UNE_153010.txt"))
    out = ns.run(dialogues_srt, frames)
    return {"narrative_text": out.narrative_text, "srt_text": out.srt_text, "approved": out.approved, "critic_feedback": out.critic_feedback}

if __name__ == "__main__":
    uvicorn.run(app, host="0.0.0.0", port=7860)