Prompt Engineering Secrets: Text‑to‑Image, Article & Code Generation with AI

Prompt engineering for text‑to‑image generation

Begin with a natural‑language description of the desired picture (e.g., a young woman in a dress watching the sunset on a beach). Use ChatGPT to rewrite the description in fluent English, then compress it into a short list of keyword tokens such as black hair, hand before body, no hand. Add negative tokens (e.g., bad hands, morbid) to suppress common artefacts. Incorporate ControlNet extensions to enforce skeletal pose, depth‑guided drawing and precise line work, which dramatically improves consistency and reduces random pose generation.

positive prompt: women back view, flowing dress, edge of the sea, back turned to the camera, sunset glow, black hair, bodycon dress
negative prompt: bad hands, morbid, lowres, blurry, watermark

Iterate the prompt, adjusting token order and weighting, and optionally use in‑painting to repair defective regions.

Prompting ChatGPT for article creation

Structure the request with the STAR (Situation‑Task‑Action‑Result) framework to guide the model toward a coherent outline. Example prompt (Chinese) that asks for a 100‑word introduction, a case study using Julia DataFrames, and a concluding summary:

我想写一篇文章，主题是《如何用ChatGPT写一篇文章，以如何用ChatGPT写一篇文章》
1. 先用100字介绍ChatGPT
2. 用Julia Dataframe在大数据分析领域的使用…
3. 总结并强调好处
帮我写一篇文章？

Providing this explicit context block yields more accurate and well‑structured output.

Domain‑Driven Design API example (TicketOrder aggregate)

The following typical CRUD and reservation operations illustrate how AI can suggest domain‑specific APIs: createOrder() – create a new order with customer, show‑time and seat data. cancelOrder(orderId) – cancel an existing order. getOrder(orderId) – retrieve order details. reserveSeats(showTimeId, seatIds) – reserve specific seats for a show. confirmOrder(orderId) – lock seats and mark the order as confirmed. payOrder(orderId, paymentInfo) – process payment. refundOrder(orderId) – issue a refund when allowed. updateOrder(orderId, updateInfo) – modify order information.

Code generation with GitHub Copilot

Copilot (released August 2021) can generate code from comments. In a Kotlin Spring controller, a comment such as @PostMapping together with a request‑body annotation leads Copilot to produce a skeleton method. Adding a detailed comment improves correctness.

@PostMapping
fun createTicketOrder(@RequestBody ticketOrder: TicketOrder): String {
    // check the schedule is valid, the seats are available, then create the order
}

When the function name and parameters are in English, Copilot’s suggestions become more accurate, and it can also generate related exception handling code or unit tests.

Combining DDD, ChatGPT, and Copilot

A practical workflow:

Use ChatGPT to design domain models and write a concise API specification.

Feed the specification to Copilot (via comments) to generate boiler‑plate code in the target language.

Convert the generated API list into a markdown table for documentation.

Model selection and safety considerations

High‑quality image results depend on the diffusion model. Model repositories such as https://civitai.com/ provide many community‑trained checkpoints. Some models contain adult or otherwise sensitive content; users should access them responsibly and avoid using them in public environments.