From Requirements to a Runnable Short-Video Prototype: An Intelligent Programming Workflow
This article uses an image-to-commerce-video product as an example to explain how to break a long-cycle software project into verifiable stages and keep AI coding tools producing consistent code under a shared specification. It covers the prerequisites for reproduction, a recommended technology stack, acceptance criteria, and why requirements should be defined before code generation. It makes no promises about development time, which depends on team experience, project scope, and the stability of third-party video APIs.
Many teams want to use AI coding tools to build complete products. A common result, however, is an application whose pages open successfully but whose workflows do not withstand careful review. Using an “image-to-commerce-video” product as an example, this article explains how to divide a long-cycle project into verifiable stages so that AI coding tools can continuously produce consistent code under a unified specification.
This article does not promise a specific development timeline. Actual effort depends on team experience, project scope, and the stability of third-party video-generation APIs.
1. Define the Prerequisites Before Reproduction
Before reproducing this type of implementation, clarify four areas:
- Team members and responsibilities
- The AI coding tool to be used
- The target runtime environment
- The acceptance criteria
The following configuration can serve as a reference. Teams can replace individual components based on their own requirements.
- Frontend: React 18, TypeScript 5, and Vite 5
- Backend: Node.js 20 LTS, NestJS 10, with Prisma 5 as the data-access layer
- Database: PostgreSQL 16, using the default
READ COMMITTEDisolation level for business transactions together with row-level locking - Queue and cache: Redis 7, with BullMQ 4 for asynchronous tasks
- Object storage: An S3-compatible service; MinIO can be used in development
- Deployment: Docker Compose for development and staging environments, with Nginx handling reverse proxying and upload-size limits
- External capabilities: A third-party video-generation API, with its request limits, timeout settings, and per-request cost recorded
Acceptance criteria can be defined in three parts:
- The primary workflow runs end to end in the test environment.
- Automated tests cover permissions, idempotency, billing, and resource release after failures.
- Core metrics have defined collection methods and measurement standards.
Only when all three conditions are met should the prototype be considered ready for review.
2. Define the Specification Before Generating Code
Directly asking an AI coding tool to “build a video-generation website” typically produces a visually complete interface, while the engineering…
