autoclip

Automatically downloads videos from YouTube and Bilibili platforms using dedicated API modules (backend.api.v1.youtube and backend.api.v1.bilibili) that handle platform-specific authentication, URL parsing, and video format selection. The system abstracts platform differences behind a unified video ingestion interface, storing downloaded content in a standardized format for downstream processing. Supports both direct URL input and account-based authentication for platform-specific features.

Extracts structured outlines from video content by feeding transcripts or visual keyframes to DashScope API (Alibaba's LLM service), generating hierarchical topic breakdowns with timestamps. The pipeline step (backend.pipeline.step1_outline) uses prompt engineering to convert unstructured video content into machine-readable outlines that segment the video into logical sections. This structured outline becomes the foundation for all downstream analysis, enabling timeline analysis and highlight detection.

Exposes all system functionality through a RESTful API built with FastAPI (backend/main.py and backend/api/v1/) with automatic OpenAPI documentation. Provides endpoints for project CRUD operations, video download/processing, clip retrieval, and status monitoring. Uses FastAPI's dependency injection for authentication, validation, and error handling. Implements proper HTTP status codes, error responses, and request/response schemas with Pydantic validation.

Implements internationalization (i18n) infrastructure supporting English and Chinese languages across frontend and backend. Frontend uses i18n library for dynamic language switching with locale-specific formatting. Backend provides language-specific API responses and LLM prompts. Documentation is maintained in both languages with synchronization mechanisms. Enables global user base without requiring separate deployments.

Provides Docker configuration for containerized deployment of the entire system (frontend, backend, Celery workers, Redis). Includes Dockerfile for building application images, docker-compose for local development with all services, and deployment guidance for production environments. Enables consistent deployment across development, staging, and production with minimal configuration drift.

Analyzes structured outlines from step 1 to create fine-grained timeline segments with topic labels and temporal boundaries (backend.pipeline.step2_timeline). Uses LLM-powered analysis to detect topic transitions, segment boundaries, and content coherence across the video duration. Produces a timeline data structure that maps each second of video to its corresponding topic, enabling precise highlight detection and clip generation downstream.

Scores video segments for highlight potential using LLM analysis (backend.pipeline.step3_scoring) that evaluates engagement, information density, emotional impact, and viewer interest signals. Assigns numerical scores to each timeline segment indicating likelihood of being a good highlight clip. Uses multi-dimensional scoring criteria (entertainment value, educational value, emotional peaks, etc.) to rank segments, enabling intelligent selection of top-N highlights without manual review.

Generates actual video clip files from scored segments using FFmpeg operations orchestrated through backend.services.video_service. Handles video codec selection, bitrate optimization, format conversion (MP4, WebM, etc.), and audio track management. Implements efficient frame-accurate clipping by calculating exact seek positions and duration parameters, avoiding re-encoding when possible to minimize processing time. Supports batch clip generation with parallel FFmpeg processes.

Manages the entire video processing pipeline as a series of asynchronous tasks using Celery (backend.core.celery_app) with Redis as the message broker. Each pipeline step (outline extraction, timeline analysis, scoring, clipping) is a separate Celery task that can be distributed across multiple worker processes. Implements task chaining to ensure steps execute in correct order, with intermediate results persisted to database. Provides real-time progress tracking and error handling with automatic retries for transient failures.

Provides real-time progress tracking for video processing operations through WebSocket connections that push status updates to the frontend as pipeline steps complete. The backend tracks task state in Redis and broadcasts progress events (step completed, percentage done, current operation) to connected clients. Frontend (frontend/src/pages/ProjectDetailPage.tsx) displays live progress bars and status messages without requiring polling. Enables users to monitor long-running operations without page refreshes.

Organizes video processing as discrete projects (backend.api.v1.projects) with full CRUD operations, metadata storage, and result persistence. Each project encapsulates a single video's processing state, including downloaded video, generated clips, processing logs, and user-defined settings. Projects are stored in database with relationships to all generated artifacts. Enables users to manage multiple videos simultaneously, revisit past processing results, and adjust parameters for re-processing.

Automatically groups generated clips into thematic collections based on topic similarity and scoring patterns (backend.pipeline.step5_collection). Uses LLM analysis to identify natural groupings of related clips and suggest collection themes. Produces curated clip sets that tell coherent stories or cover specific topics, rather than just ranked individual clips. Enables users to publish clip collections as compilations or playlists.

Provides a responsive web interface (frontend/src/) built with React 18+ for managing projects, uploading videos, monitoring progress, and previewing generated clips. Key components include HomePage for project listing/creation, ProjectDetailPage for real-time progress monitoring, UploadModal for video input, and ClipCard for individual clip preview and management. Uses centralized API client (frontend/src/services/api.ts) with TypeScript for type safety. Implements responsive design for desktop and mobile viewing.