witty-suckerpunch492/daihuo-jianshou

Daihuo Jianshou (loosely: "product promotion video editor") is a web application for e-commerce sellers who need to produce short promotional videos at scale for platforms like Douyin (TikTok's Chinese counterpart), Kuaishou, and Xiaohongshu (a lifestyle platform similar to Instagram). The core workflow is straightforward: upload product images, let AI generate a marketing script and individual visual assets, then automatically assemble them into a finished video ready to publish. The AI script generation draws on templates organized by product category (skincare, food, home goods, clothing, and electronics) and writing style (problem-based, scene-based, comparison, or story-driven). It also draws on a small library of proven opening-hook strategies, such as visual shock, suspenseful questions, or side by side comparisons, meant to catch a viewer's attention in the first few seconds. It also produces platform-specific metadata like hashtags, cover text, and engagement prompts tuned to the algorithm conventions of each target platform. For the visual production, the tool integrates with multiple AI image and video generation services simultaneously, giving sellers access to over twenty models from five different providers through a single interface, so a seller is not locked into one vendor's pricing or quality. Static product photos can be animated using AI-generated camera motion effects so they feel more dynamic on screen, and the README also describes four shooting styles: close-up product shots, mixed image and text cuts, staged usage scenes, and real person appearances using an uploaded likeness. All clips are assembled into a finished video using FFmpeg, with transitions, subtitles, background music mixing, and automatic aspect ratio adaptation for each platform's format. Additional features include a product library so item details only need to be entered once, batch generation for producing many product videos in one run, a template system for reusing successful scripts across new products, brand settings for consistent watermarks and colors, character management so a spokesperson can be reused across projects, and A/B variant generation to test multiple opening lines or scripts. The application is built with Next.js, React, and TypeScript, and stores its data in a local SQLite database. It is released under the MIT license.