Mux audio and video.

Função para substituir texto.

Crops the center of the frame to fill 9:16 aspect ratio (Zoom effect).

Returns (width, height, duration_frames)

Recorta uma região baseada no rosto mantendo o aspect ratio do target. Previne deformação (esticar/espremer).

Recorta e redimensiona dois rostos detectados no frame, ajustando para uma composição vertical 1080x1920 onde cada rosto ocupa metade da tela

Detect faces using InsightFace. Returns a list of dicts with 'bbox' and 'kps'. bbox is [x1, y1, x2, y2], kps is 5 keypoints (eyes, nose,

Fallback function: Center Crop (Zoom) or Padding if detection fails.

Returns the subtitle configuration dictionary. Can be expanded to load from a JSON/YAML file in the future.

Explicit initialization if needed outside import.

Sorts new_faces to match the order of old_faces based on distance. new_faces: list of face objects (bbox or tuple) old_faces: list of

Workaround for 'Weights only load failed' error in newer PyTorch versions. We first try to add safe globals. If that's not enough/fails, we

Burns subtitles into a single video file.

Calculate Mouth Aspect Ratio (MAR) using 68-point landmarks (inner lips). Indices: Inner Lips: 60-67 (0-indexed 60 to 67) Left C

Limpa a resposta focando em encontrar o objeto JSON que contém a chave "segments". Estratégia: 1. Busca a palavra "segments", encontr

Generates a Premiere Pro XML with segmented cuts, supporting Dual-Track (Split Screen) for multi-face scenarios.

Crops and resizes the frame to target dimensions centered on the face_bbox. face_bbox: [x1, y1, x2, y2]

Runs face detection on the fly if pre-computed coords are missing. Returns: list of {'frame': int, 'faces': [[x1,y1,x2,y2]]}

Generates a ZIP Pack for the segment.

Generates a single ASS file from a JSON input.

Face detection using OpenCV Haar Cascades.

Face detection using InsightFace (SOTA).

Solicita um inteiro ao usuário via terminal.

Converts internal JSON subtitle format to SRT. If 'words' key is present, generates word-level timestamps (Karaoke/Editing style). Otherw

Parses input.tsv or input.srt from the project folder.

Parses an SRT file into a list of segments expected by WhisperX alignment. [{'start': float, 'end': float, 'text': str}, ...]

Parses a VTT file (WebVTT) into valid segments for WhisperX.

Concatenates transcript segments into a single string with embedded time tags.

Aligns raw AI segments (with reference tags) to actual transcript timestamps. Applies constraints, validation, and deduplication.

Renders segments using a physical transparent PNG canvas to ensure alpha correctness.

Remove caracteres inválidos e emojis para evitar erro de encoding no Windows.

Remove caracteres inválidos para nomes de arquivos/pastas.

A context manager that redirects stdout and stderr to devnull

Generates a Premiere Pro XML (xmeml version 4) compliant with vertical restrictions. video_path: Path to the main video (Track 1) overlay

Lê o input.json (WhisperX), recorta o trecho e salva em output_json_path com timestamps ajustados.

Generates HTML gallery for a given project folder using FastAPI Static Files mounting.

Scans VIRALS/{project_name}/subs/ for json files.