KryinCaption AI: 9-Step Agentic Video Director

KryinCaption AI is a state-of-the-art 9-step agentic video understanding and captioning system engineered to analyze short video clips with high speed and precision. Unlike traditional captioners that process videos as flat sequences, KryinCaption uses a dual-agent architecture modeled after a creative production studio: a Writer Agent drafts captions, and an Editorial Director Agent critiques and refines them. The pipeline runs through 9 distinct steps: 1. Video Download & Validation (supporting direct files and YouTube/Shorts). 2. Zone-Based Frame Extraction (sampling 6 frames across beginning, middle, and ending temporal zones to capture the full narrative arc). 3. Audio Extraction. 4. Speech Transcription (using Whisper with custom hallucination filtering). 5. Scene Intelligence Analysis (generating mood, emotional hooks, story arcs, and attention grabbers). 6. Caption Strategy Construction. 7. Multi-Style Caption Generation (generating Formal, Sarcastic, Tech Humor, Non-Tech Humor, and Social Media variations). 8. Creative Director Review (scoring captions on style authenticity, human-likeness, creativity, and style strength). 9. Iterative Refinement (automatically rewriting and re-grading underperforming captions). Optimized for high-performance open weights models, the pipeline defaults to serverless GLM 5.2 for writing and Gemma 4 for criticism, with automated, zero-latency multi-tier fallbacks to Google Gemini Flash/Pro to guarantee absolute system reliability. Ready for containerized deployment, it features a glassmorphic web dashboard and a parallel CLI batch processor.