aitx-codex-hackathon/reel_studio

ReelStudio is a Python application that turns a folder of real estate listing photos into a professionally sequenced, beat-synchronised property video reel, requiring no video editing experience. It targets real estate agents, property managers, and marketing teams who need polished listing videos without hiring a videographer. The workflow is a four-step wizard: upload photos, pick a music track from the built-in catalogue, review a generated storyboard, then render the final video. Under the hood, a vision AI powered by GPT-4o analyses each photo to classify it by room type, a storyboard agent orders the shots in a fixed cinematic house-tour arc (exterior to foyer to kitchen to living room to bedroom to bathroom to backyard), and FAL's Kling video generation model produces five-second cinematic clips with realistic camera motion for each shot. FFmpeg stitches those clips into a beat-synchronised 9:16 MP4 ready for Instagram Reels, TikTok, or MLS embeds. The system accepts between 3 and 150 photos and costs roughly $0.40 to $0.80 per render in API credits. A two-pass render system first produces a fast 540p draft preview, then a full 1080p final. Beat timing is extracted from the music track using librosa so cut points land on musical beats before any video is generated. Hallucination guardrails in every FAL prompt explicitly instruct the model to preserve real architecture and not invent new rooms, important to avoid legal liability in property listings. The backend uses FastAPI with SQLite for storage and streams live progress to the browser via WebSockets. The frontend is a Next.js 14 wizard with Tailwind. The full README is longer than what was provided.