从零开始架构了完整的 NestJS 后端——包括 Mongoose schema、模块化 CRUD API、DTO 验证,以及将前端扁平数组转换为后端嵌入式文档的映射层。
构建了组合计算器引擎,它枚举所有有效的 Hook x Body x CTA x Subtitle 组合,并将每个组合作为一个独立的 BullMQ 作业入队。
开发了具有 6 种动画类型的 Remotion 组合(VideoComposition, HookOverlay, BodyOverlay, CtaOverlay),并在服务器启动时进行捆绑,以实现快速作业执行。
实现了 BullMQ 生成处理器:S3 下载、Remotion renderMedia()、FFmpeg 编码优化、缩略图生成、S3 上传和数据库更新,并包含错误处理和重试机制。
在 SubtitleRenderer.tsx 中构建了 6 种字幕渲染风格,涵盖了所有 4 种宽高比下的 Hormozi、Abdal、MrBeast、GaryVee、TikTok Viral 和 Clean Corporate 风格。
集成了 WebSocket 网关,实现实时生成进度,并带有 SSE 备用方案,无需轮询即可在前端实现实时作业状态更新。
实现了完整的 Stripe 订阅生命周期——包括结账会话、账单门户、带有原始主体和签名验证的 webhook 处理器、订阅时的信用分配以及每月信用重置 cron。
构建了分享与下载系统——基于 nanoid 的可分享令牌、无需身份验证的公共 S3 签名 URL 访问、链接撤销以及 Content-Disposition 下载头。
开发了管理模块,利用 MongoDB 聚合管道实现仪表盘分析、时间序列图表数据、用户管理和订阅管理。
主导了用户应用和管理面板的前后端全面集成——用实时 API 调用替换了所有模拟存储,连接了 WebSocket 进度,并端到端地验证了完整的用户旅程。
通过自动化组合生成、GPU 加速渲染和多风格字幕支持,将单个视频上传转换为数百种品牌广告变体——专为效果营销人员打造。






MicrocosmWorks built the combination calculator engine that computes the Cartesian product of all Hook x Body x CTA x Subtitle permutations. Each input element (hooks, body segments, CTAs) is defined by the user, and the engine enumerates every valid combination, then enqueues each as an independent BullMQ job. This approach means a video with 5 hooks, 3 bodies, 4 CTAs, and 6 subtitle styles generates 360 unique ad variations automatically.
MicrocosmWorks implemented 6 subtitle rendering styles in Adstacker: Hormozi, MrBeast, GaryVee, Abdal, TikTok Viral, and Clean Corporate. Each style has distinct font, color, animation, and positioning characteristics matching the popular creator formats that performance marketers recognize. The SubtitleRenderer handles all 4 aspect ratios (9:16, 16:9, 1:1, 4:5) and 6 animation types (fade, slide-up, slide-down, scale, typewriter, bounce) within Remotion compositions.
MicrocosmWorks built the rendering pipeline using Remotion compositions bundled at server startup for fast job execution, combined with FFmpeg for encoding optimization. BullMQ manages the async job queue with configurable concurrency control, so multiple variations render in parallel across available GPU resources. Each job handles the full pipeline: S3 download, Remotion renderMedia, FFmpeg encoding, thumbnail generation, S3 upload, and database update, with error handling and automatic retries.
MicrocosmWorks integrated a WebSocket gateway that pushes real-time generation progress for every queued variation to the frontend, with SSE (Server-Sent Events) fallback for environments where WebSocket connections are unavailable. Users see live status updates for each variation including queued, processing, rendering percentage, and completed states. This eliminates the need for polling and provides immediate feedback on the generation pipeline status.
MicrocosmWorks implemented Stripe subscription billing with webhook-driven credit allocation. When a user subscribes or renews, the webhook handler allocates generation credits based on their plan tier. Each video variation generation consumes one credit, and a monthly cron job resets credit balances. The admin dashboard provides MongoDB aggregation analytics showing credit usage patterns, subscription revenue, and per-user consumption to inform pricing decisions.