pipeline/processor.py

"""
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  SUPREMEAI — Processeur Vidéo Avancé
  • Post-processing : upscale, interpolation, color grading
  • Backend CPU : génération vidéo sans GPU (tutoriels, motion design)
  • Super-résolution via Real-ESRGAN
  • Interpolation frames via RIFE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
"""

import os, time, math, io, re
from pathlib import Path
from typing import Optional, List
import numpy as np

from PIL import Image, ImageDraw, ImageFont, ImageFilter, ImageEnhance, ImageOps
import cv2

OUTPUT_DIR = Path(os.getenv("SUPREMEAI_OUTPUT", "/tmp/supremeai_output"))
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)

# ── LUTs de color grading ────────────────────────────────────────────────────
def apply_lut_cinematic(frame: np.ndarray) -> np.ndarray:
    """Applique un LUT cinématique (tons chauds, contraste élevé)."""
    # Courbe S pour le contraste
    lut = np.arange(256, dtype=np.float32)
    lut = 255 * (1 / (1 + np.exp(-0.05 * (lut - 128))))
    lut = lut.astype(np.uint8)
    # Boost rouge/jaune, légère désaturation bleu
    result = frame.copy()
    result[:, :, 2] = cv2.LUT(frame[:, :, 2], lut)         # Rouge
    result[:, :, 1] = cv2.LUT(frame[:, :, 1], (lut * 0.95).astype(np.uint8))  # Vert
    result[:, :, 0] = cv2.LUT(frame[:, :, 0], (lut * 0.85).astype(np.uint8))  # Bleu
    return result

def apply_lut_warm(frame: np.ndarray) -> np.ndarray:
    """LUT chaud — style golden hour."""
    result = frame.copy().astype(np.float32)
    result[:, :, 2] = np.clip(result[:, :, 2] * 1.10, 0, 255)
    result[:, :, 1] = np.clip(result[:, :, 1] * 1.02, 0, 255)
    result[:, :, 0] = np.clip(result[:, :, 0] * 0.90, 0, 255)
    return result.astype(np.uint8)

def apply_lut_cold(frame: np.ndarray) -> np.ndarray:
    """LUT froid — style sci-fi/futuriste."""
    result = frame.copy().astype(np.float32)
    result[:, :, 2] = np.clip(result[:, :, 2] * 0.90, 0, 255)
    result[:, :, 1] = np.clip(result[:, :, 1] * 1.02, 0, 255)
    result[:, :, 0] = np.clip(result[:, :, 0] * 1.15, 0, 255)
    return result.astype(np.uint8)

def apply_lut_vintage(frame: np.ndarray) -> np.ndarray:
    """LUT vintage — désaturation + grain."""
    gray  = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    sepia = cv2.merge([
        np.clip(gray * 0.68, 0, 255).astype(np.uint8),
        np.clip(gray * 0.84, 0, 255).astype(np.uint8),
        np.clip(gray * 1.06, 0, 255).astype(np.uint8),
    ])
    # Mélange avec original
    return cv2.addWeighted(frame, 0.3, sepia, 0.7, 0)

LUT_MAP = {
    "cinematic": apply_lut_cinematic,
    "warm":      apply_lut_warm,
    "cold":      apply_lut_cold,
    "vintage":   apply_lut_vintage,
    "none":      lambda f: f,
}


class FrameInterpolator:
    """Interpolation de frames avec RIFE (Real-Time Intermediate Flow Estimation).
    Permet de passer de 24fps → 60fps ou 120fps."""

    @staticmethod
    def interpolate_frames(frames: List[np.ndarray], target_fps: int, source_fps: int) -> List[np.ndarray]:
        """Interpole des frames pour augmenter le FPS."""
        if target_fps <= source_fps:
            return frames

        multiplier = target_fps // source_fps
        result = []
        for i in range(len(frames) - 1):
            result.append(frames[i])
            f0 = frames[i].astype(np.float32)
            f1 = frames[i + 1].astype(np.float32)
            for j in range(1, multiplier):
                alpha = j / multiplier
                interpolated = cv2.addWeighted(f0, 1 - alpha, f1, alpha, 0)
                result.append(interpolated.astype(np.uint8))
        result.append(frames[-1])
        return result


class SuperResolutionUpscaler:
    """Upscaling via Real-ESRGAN ou bicubique selon disponibilité."""

    def __init__(self):
        self._model = None
        self._try_load_esrgan()

    def _try_load_esrgan(self):
        try:
            from basicsr.archs.rrdbnet_arch import RRDBNet
            from realesrgan import RealESRGANer
            model = RRDBNet(num_in_ch=3, num_out_ch=3, num_feat=64,
                            num_block=23, num_grow_ch=32, scale=4)
            self._model = RealESRGANer(
                scale=4, model_path="weights/RealESRGAN_x4plus.pth",
                model=model, tile=256, tile_pad=10, pre_pad=0, half=True,
            )
        except Exception:
            self._model = None  # fallback bicubique

    def upscale(self, frame: np.ndarray, scale: int = 2) -> np.ndarray:
        if self._model is not None:
            try:
                output, _ = self._model.enhance(frame, outscale=scale)
                return output
            except Exception:
                pass
        # Fallback : bicubique (qualité correcte sans GPU dédié)
        h, w = frame.shape[:2]
        return cv2.resize(frame, (w * scale, h * scale), interpolation=cv2.INTER_CUBIC)


class MotionEffectsEngine:
    """Effets de mouvement : Ken Burns, zoom cinématique, transitions."""

    @staticmethod
    def ken_burns(frame: np.ndarray, progress: float, zoom_start: float = 1.0,
                  zoom_end: float = 1.15, pan_x: float = 0.0, pan_y: float = 0.0) -> np.ndarray:
        """Effet Ken Burns : zoom progressif + panoramique."""
        h, w = frame.shape[:2]
        zoom = zoom_start + (zoom_end - zoom_start) * progress
        new_w, new_h = int(w * zoom), int(h * zoom)
        resized = cv2.resize(frame, (new_w, new_h), interpolation=cv2.INTER_LANCZOS4)
        # Crop centré avec décalage panoramique
        x0 = (new_w - w) // 2 + int(pan_x * progress * w * 0.1)
        y0 = (new_h - h) // 2 + int(pan_y * progress * h * 0.1)
        x0 = max(0, min(x0, new_w - w))
        y0 = max(0, min(y0, new_h - h))
        return resized[y0:y0+h, x0:x0+w]

    @staticmethod
    def transition_fade(f1: np.ndarray, f2: np.ndarray, alpha: float) -> np.ndarray:
        return cv2.addWeighted(f1, 1 - alpha, f2, alpha, 0)

    @staticmethod
    def transition_wipe_left(f1: np.ndarray, f2: np.ndarray, alpha: float) -> np.ndarray:
        w = f1.shape[1]
        split = int(w * alpha)
        result = f1.copy()
        result[:, :split] = f2[:, :split]
        return result

    @staticmethod
    def transition_zoom_out(f1: np.ndarray, f2: np.ndarray, alpha: float) -> np.ndarray:
        h, w = f1.shape[:2]
        scale = 1.0 - alpha * 0.3
        new_w, new_h = int(w * scale), int(h * scale)
        small = cv2.resize(f1, (new_w, new_h), interpolation=cv2.INTER_LINEAR)
        result = f2.copy()
        y0 = (h - new_h) // 2
        x0 = (w - new_w) // 2
        result[y0:y0+new_h, x0:x0+new_w] = small
        return result

    TRANSITIONS = {
        "fade":      transition_fade.__func__,
        "wipe_left": transition_wipe_left.__func__,
        "zoom_out":  transition_zoom_out.__func__,
    }


class TextRenderer:
    """Rendu de texte avancé avec ombre, glow, outline."""

    @staticmethod
    def _get_font(size: int, bold: bool = False) -> ImageFont.FreeTypeFont:
        candidates = [
            "/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf",
            "/usr/share/fonts/truetype/liberation/LiberationSans-Bold.ttf",
            "/usr/share/fonts/truetype/ubuntu/Ubuntu-B.ttf",
        ]
        for path in candidates:
            if os.path.exists(path):
                try:
                    return ImageFont.truetype(path, size)
                except Exception:
                    continue
        return ImageFont.load_default()

    @classmethod
    def draw_text_with_effects(
        cls, img: Image.Image, text: str, x: int, y: int,
        font_size: int = 40, color: tuple = (255, 255, 255),
        shadow: bool = True, outline: bool = False, glow: bool = False,
        align: str = "left"
    ) -> Image.Image:
        font = cls._get_font(font_size)
        draw = ImageDraw.Draw(img)

        # Découpe le texte en lignes si trop long
        words  = text.split()
        lines  = []
        cur    = ""
        max_w  = img.width - x - 20
        for word in words:
            test = (cur + " " + word).strip()
            bbox = draw.textbbox((0, 0), test, font=font)
            if bbox[2] <= max_w:
                cur = test
            else:
                if cur:
                    lines.append(cur)
                cur = word
        if cur:
            lines.append(cur)

        line_h = font_size + 6
        for i, line in enumerate(lines):
            ly = y + i * line_h
            bbox = draw.textbbox((0, 0), line, font=font)
            lw = bbox[2] - bbox[0]

            if align == "center":
                lx = (img.width - lw) // 2
            elif align == "right":
                lx = img.width - lw - x
            else:
                lx = x

            if glow:
                # Effet glow : dessine le texte en blanc flou
                glow_img = Image.new("RGBA", img.size, (0, 0, 0, 0))
                gd = ImageDraw.Draw(glow_img)
                for dx in range(-3, 4):
                    for dy in range(-3, 4):
                        gd.text((lx+dx, ly+dy), line, font=font, fill=(255, 255, 200, 80))
                glow_img = glow_img.filter(ImageFilter.GaussianBlur(3))
                if img.mode != "RGBA":
                    img = img.convert("RGBA")
                img = Image.alpha_composite(img, glow_img)
                draw = ImageDraw.Draw(img)

            if shadow:
                draw.text((lx+2, ly+2), line, font=font, fill=(0, 0, 0, 180))

            if outline:
                for ox, oy in [(-1,-1),(1,-1),(-1,1),(1,1)]:
                    draw.text((lx+ox, ly+oy), line, font=font, fill=(0, 0, 0))

            draw.text((lx, ly), line, font=font, fill=color)

        return img


class EnhancedVideoProcessor:
    """
    Générateur vidéo CPU professionnel.
    Crée des vidéos de qualité broadcast sans GPU.
    Utilisé pour les tutoriels, motion design, éducatif.
    """

    THEMES = {
        "dark_gold":   {"bg": (10, 22, 40), "accent": (201, 168, 76), "text": (220, 232, 255)},
        "dark_purple": {"bg": (20, 10, 35), "accent": (180, 80, 255), "text": (230, 220, 255)},
        "dark_red":    {"bg": (25, 8, 8),   "accent": (255, 60, 60),  "text": (255, 230, 230)},
        "dark_blue":   {"bg": (8, 15, 35),  "accent": (60, 140, 255), "text": (220, 235, 255)},
        "dark_green":  {"bg": (8, 25, 15),  "accent": (60, 200, 120), "text": (220, 255, 230)},
        "minimal_light": {"bg": (245, 245, 250), "accent": (30, 100, 200), "text": (20, 20, 40)},
    }

    def __init__(self, config):
        self.cfg       = config
        self.W         = config.width
        self.H         = config.height
        self.fps       = config.fps
        self.duration  = config.duration
        self.theme     = self.THEMES.get("dark_gold")
        self.motion    = MotionEffectsEngine()
        self.tr        = TextRenderer()

    def _make_gradient_bg(self, frame_idx: int, total_frames: int) -> Image.Image:
        """Fond dégradé animé avec particules."""
        bg = self.theme["bg"]
        acc = self.theme["accent"]
        progress = frame_idx / max(total_frames - 1, 1)

        img = Image.new("RGB", (self.W, self.H), bg)
        draw = ImageDraw.Draw(img)

        # Dégradé radial animé
        cx = int(self.W * (0.3 + 0.1 * math.sin(progress * 2 * math.pi)))
        cy = int(self.H * (0.4 + 0.05 * math.cos(progress * 2 * math.pi)))
        for r in range(min(self.W, self.H) // 2, 0, -2):
            alpha = max(0, 1 - r / (min(self.W, self.H) * 0.8))
            color = tuple(int(bg[c] + (acc[c] - bg[c]) * alpha * 0.25) for c in range(3))
            draw.ellipse([cx-r, cy-r, cx+r, cy+r], fill=color)

        # Particules flottantes
        rng = np.random.default_rng(frame_idx // 3)
        for _ in range(12):
            px = rng.integers(0, self.W)
            py = int((rng.integers(0, self.H) + frame_idx * 0.5) % self.H)
            pr = rng.integers(1, 4)
            pa = int(rng.integers(40, 100))
            color = tuple(min(255, int(c * 1.5)) for c in acc)
            draw.ellipse([px-pr, py-pr, px+pr, py+pr], fill=color)

        return img

    def _make_scene_frame(self, text: str, frame_idx: int, total_frames: int,
                          scene_idx: int, n_scenes: int, title: str = "") -> Image.Image:
        """Crée un frame avec fond animé, titre et texte."""
        img = self._make_gradient_bg(frame_idx, total_frames)
        acc = self.theme["accent"]
        txt = self.theme["text"]
        bg  = self.theme["bg"]
        progress = frame_idx / max(total_frames - 1, 1)

        # Barre de progression en bas
        bar_h = 4
        bar_w = int(self.W * ((scene_idx + progress) / n_scenes))
        draw  = ImageDraw.Draw(img)
        draw.rectangle([0, self.H - bar_h, self.W, self.H], fill=tuple(int(c * 0.3) for c in acc))
        draw.rectangle([0, self.H - bar_h, bar_w, self.H], fill=acc)

        # Ligne décorative en haut
        draw.rectangle([0, 0, self.W, 3], fill=acc)
        draw.rectangle([0, 4, int(self.W * 0.6), 6], fill=tuple(int(c * 0.5) for c in acc))

        # Numéro scène / logo
        draw.text((self.W - 120, 12), f"⬡ {scene_idx+1}/{n_scenes}", fill=acc,
                  font=TextRenderer._get_font(18))

        # Slide-in animé pour le texte
        slide_progress = min(1.0, progress * 6)  # slide in rapide
        slide_x = int((1 - slide_progress) * self.W * 0.3)

        if title:
            # Titre principal centré
            title_y = int(self.H * 0.30)
            img = self.tr.draw_text_with_effects(
                img, title, slide_x + 40, title_y,
                font_size=min(56, max(28, int(280 / max(1, len(title))))),
                color=acc, shadow=True, glow=True, align="center"
            )

        # Corps texte
        text_y = int(self.H * (0.50 if title else 0.35))
        img = self.tr.draw_text_with_effects(
            img, text, slide_x + 40, text_y,
            font_size=min(38, max(20, int(180 / max(1, len(text) // 3)))),
            color=txt, shadow=True, outline=False, align="center"
        )

        # Logo watermark
        draw = ImageDraw.Draw(img)
        draw.text((10, self.H - 24), "SupremeAI · Informa-Technique R",
                  fill=tuple(int(c * 0.4) for c in txt),
                  font=TextRenderer._get_font(14))

        return img

    def render(self) -> Path:
        """Rendu complet de la vidéo."""
        from moviepy import VideoClip, VideoFileClip, concatenate_videoclips
        import tempfile

        prompt    = self.cfg.prompt
        total_sec = self.cfg.duration
        total_frames = int(total_sec * self.fps)

        # Découpe le prompt en scènes
        sentences = [s.strip() for s in re.split(r'[.!?\n]+', prompt) if len(s.strip()) > 5]
        if not sentences:
            sentences = [prompt]
        n_scenes = len(sentences)

        # Génère les frames
        all_frames = []
        frames_per_scene = max(1, total_frames // n_scenes)

        for si, sentence in enumerate(sentences):
            # Titre = première scène en gros
            title = sentences[0][:60] if si == 0 else ""
            is_last = (si == n_scenes - 1)

            for fi in range(frames_per_scene):
                frame = self._make_scene_frame(
                    text=sentence, frame_idx=fi,
                    total_frames=frames_per_scene,
                    scene_idx=si, n_scenes=n_scenes,
                    title=title if fi < frames_per_scene // 3 else "",
                )
                # Ken Burns sur la frame
                frame_np = np.array(frame)
                prog = fi / max(frames_per_scene - 1, 1)
                frame_np = self.motion.ken_burns(frame_np, prog,
                    zoom_start=1.0, zoom_end=1.08,
                    pan_x=(si % 2) * 0.5, pan_y=0.0)

                # Color grading
                lut_fn = LUT_MAP.get(self.cfg.color_grading, LUT_MAP["none"])
                frame_np = lut_fn(frame_np)

                all_frames.append(frame_np)

        # Transition fade entre scènes
        if len(all_frames) >= frames_per_scene * 2:
            trans_frames = int(self.fps * 0.3)  # 0.3 seconde de transition
            final_frames = []
            for si in range(n_scenes - 1):
                scene_start = si * frames_per_scene
                scene_end   = scene_start + frames_per_scene
                next_start  = scene_end

                final_frames.extend(all_frames[scene_start:scene_end - trans_frames])
                for ti in range(trans_frames):
                    alpha = ti / trans_frames
                    blended = cv2.addWeighted(
                        all_frames[scene_end - trans_frames + ti], 1 - alpha,
                        all_frames[min(next_start + ti, len(all_frames) - 1)], alpha, 0
                    )
                    final_frames.append(blended)
            final_frames.extend(all_frames[n_scenes * frames_per_scene - frames_per_scene:])
        else:
            final_frames = all_frames

        # Interpolation FPS si demandée
        if self.cfg.interpolate_fps > self.fps:
            interp = FrameInterpolator()
            final_frames = interp.interpolate_frames(
                final_frames, self.cfg.interpolate_fps, self.fps)
            actual_fps = self.cfg.interpolate_fps
        else:
            actual_fps = self.fps

        # Upscale si demandé
        if self.cfg.upscale_to_4k:
            upscaler = SuperResolutionUpscaler()
            final_frames = [upscaler.upscale(f, scale=2) for f in final_frames]

        # Assemblage MoviePy
        def make_frame(t):
            idx = min(int(t * actual_fps), len(final_frames) - 1)
            return cv2.cvtColor(final_frames[idx], cv2.COLOR_BGR2RGB)

        clip = VideoClip(make_frame, duration=total_sec)

        out_path = OUTPUT_DIR / f"supremeai_{int(time.time())}.mp4"
        clip.write_videofile(
            str(out_path), fps=actual_fps,
            codec="libx264", audio=False,
            ffmpeg_params=["-crf", "18", "-preset", "fast"],
            logger=None,
        )
        return out_path


class AudioSyncEngine:
    """Synchronisation audio-vidéo et génération voix-off."""

    @staticmethod
    async def generate_voiceover(text: str, lang: str, output_path: str) -> str:
        """Génère une voix-off via edge-tts."""
        try:
            import edge_tts
            voices = {
                "fr": "fr-FR-DeniseNeural",
                "en": "en-US-AriaNeural",
                "ar": "ar-DZ-AminaNeural",
                "es": "es-ES-ElviraNeural",
            }
            voice = voices.get(lang, voices["fr"])
            comm  = edge_tts.Communicate(text, voice)
            await comm.save(output_path)
            return output_path
        except Exception as e:
            return ""

    @staticmethod
    def add_audio_to_video(video_path: str, audio_path: str, output_path: str,
                           music_path: Optional[str] = None,
                           music_volume: float = 0.12) -> str:
        """Combine vidéo + voix-off + musique de fond."""
        from moviepy import VideoFileClip, AudioFileClip, CompositeAudioClip
        import moviepy.audio.fx as afx

        video = VideoFileClip(video_path)
        audio_clips = []

        if audio_path and os.path.exists(audio_path):
            voice = AudioFileClip(audio_path).subclipped(0, video.duration)
            audio_clips.append(voice)

        if music_path and os.path.exists(music_path):
            music = (AudioFileClip(music_path)
                     .subclipped(0, video.duration)
                     .with_effects([afx.MultiplyVolume(music_volume)]))
            audio_clips.append(music)

        if audio_clips:
            final_audio = (CompositeAudioClip(audio_clips)
                           if len(audio_clips) > 1
                           else audio_clips[0])
            video = video.with_audio(final_audio)

        video.write_videofile(output_path, codec="libx264", audio_codec="aac", logger=None)
        return output_path