I spent ~4 hours building AIComicBuilder [1], a full-stack AI comic/drama generation platform, without writing a single line of code manually. This is a writeup of my "vibe coding" workflow using Claude Code.

  The app lets you: input a script → AI generates screenplay → character analysis → storyboard
  generation (with first/last frame images) → video generation. Supports OpenAI-compatible,
  Gemini, and Bytedance Seedance APIs for text/image/video models, configurable per-user via a
  settings UI.

  Tech: Next.js full-stack, SQLite, queue-based async processing.

  **The workflow:**
  1. Describe tech requirements in plain text → AI generates architecture doc → review/tweak in
     the terminal
  2. Reference the doc, load `superpowers:subagent-driven-development` → AI writes the plan,
     self-reviews it, then executes autonomously (~45 min)
  3. Auto-commits after each feature (critical safety net if AI breaks something)
  4. Bug? Screenshot + one sentence → AI fixes it
  5. Ugly UI? "redesign using ui-ux-pro-max" → 3 iterations → done

  The key insight from Karpathy's "vibe coding" concept: in this paradigm, *documentation quality
  matters more than coding ability*. The clearer your requirements, the better the output. Your
  role shifts from programmer to product manager.

  I'm documenting the full workflow and useful Claude Code skills here: [2]

  Curious if others have found good skill combinations for Claude Code, or are using similar
  approaches with other tools (Cursor, Aider, etc.).

  [1] https://github.com/twwch/AIComicBuilder
  [2] https://github.com/twwch/vibe-coding