Custom Prompt Templates

​

Creating Text Generation Templates

Primarily used for text creation, knowledge Q&A, code generation, and conversational interaction. Supports structured design via prompt engineering frameworks. Two creation methods are available:

• Custom Creation: Ideal when you already have a working prompt and want to quickly convert it into a template or perform simple optimizations.
• Prompt Engineering–Based Creation: Best suited for building high-quality, complex-task prompts—or when you wish to leverage professional frameworks to spark ideas and ensure structural completeness.

​

Method 1: Custom Creation

1. Navigate to Component Management > Prompts page and click + Create Prompt. Configure the following:
   • Template Name: Enter a custom name.
   • Model Type: Select Text Generation.
   • Task Type: Select a task type.
2. In the Template Content & Optimization section, enter your prompt (e.g., Please help me write a 100-word sales data analysis report) in the text input box, then click the Optimize button (top-right corner) to refine it.

Optimization Example: Enter your original prompt in the prompt editor, e.g., Please help me write a 100-word sales data analysis report, then click the Optimize button (top-right corner). The optimized prompt expands the original concise request into a structured instruction, including:

• Task Instruction: Write a concise sales data analysis report (~100 words).
• Requirement 1: Briefly summarize sales performance.
• Requirement 2: Include key metrics—total sales revenue, best-selling product, growth rate.
• Requirement 3: Analyze underlying causes behind the data and identify trends.
• Requirement 4: Conclude with actionable recommendations or next steps.

Ends with: "Please complete this sales data analysis report according to the above guidelines." Character count: 189 / 8,000.

3. After optimization, click Create, and the template is successfully created.

Upon successful creation, the page automatically redirects to the resource detail page, displaying basic configuration parameters in a two-column table layout.

​

Method 2: Prompt Engineering–Based Creation

1. Navigate to Component Management > Prompts page and click + Create Prompt. Configure the following:
   • Template Name: Enter a custom name.
   • Model Type: Select Text Generation.
   • Task Type: Select a task type.
2. Select a prompt engineering framework. Different frameworks suit tasks of varying complexity—see the reference table in Prompt Engineering Frameworks Explained.
3. Fill in the corresponding input fields based on the selected framework's structure.

Example: Using the ICIO framework, fill in the following:

• Instruction: Please write a blog post about sustainable energy (~1,000 words), suitable for intermediate-level readers.
• Context: The world currently faces an energy crisis, and interest in environmentally friendly energy solutions is growing. The article should cover fundamental concepts of sustainable energy, major types, and how they help reduce carbon footprints.
• Input Data: Consider the following data points: recent developments in solar and wind energy, government support for sustainable energy projects, and their potential impact on greenhouse gas reduction.
• Output Indicator: The article must be well-structured (introduction, main body, conclusion), written in accessible language with minimal technical jargon. Include factual data and citations to strengthen credibility.

4. Click Optimize, and the system will intelligently polish and enhance your inputs.

Key Improvements:

• Precision in Objective: Clearly defines the core educational goal and strengthens strategic alignment.
• Deepened Context: Adds causal chain (fossil fuels → crisis → demand → solution) to build a compelling narrative arc.
• Concrete Data: Names specific renewable sources (solar/wind/hydro) to guide practical application.
• Enhanced Structure: Shifts from listing elements to a cognitive journey: engaging hook → detailed exposition → action-oriented conclusion.
• Layered Language: Balances technical accuracy with accessibility, precisely calibrated for intermediate readers.
• Trust Architecture: Establishes dual verification via facts + authoritative citations.

5. After optimization, click Create, and the template is successfully created.

​

Creating Image Generation Templates

Designed primarily for image creation scenarios. Supports separate definition of positive and negative prompts to control visual content and artistic style.

1. Navigate to Component Management > Prompts page and click + Create Prompt. Configure the following:
   • Template Name: Enter a custom name.
   • Model Type: Select Image Generation.
2. Enter the Positive Prompt and Negative Prompt separately.

Positive prompts instruct the large model on desired visual elements; negative prompts specify undesired elements. Thoughtful use of both significantly improves alignment between generated images and user intent. Example:

• Positive Prompt: Generate a vibrant night sky filled with twinkling stars and streaking meteors, reflected in the calm surface of a serene lake—evoking tranquility and wonder.
• Negative Prompt: Exclude dark or blurry backgrounds; avoid black or deep gray tones. Do not include urban architecture, withered trees, or any oppressive elements. Ensure the overall mood remains bright and uplifting.

3. After entering both prompts, click Create, and the template is successfully created.

​

Prompt Engineering Frameworks Explained

The platform includes multiple built-in prompt engineering frameworks—providing structured mental models for complex tasks.

​

ICIO Framework

​

CRISPE Framework

​

RASCEF Framework

​

Optimization Examples

The following shows before/after optimization comparisons for each framework: ICIO Optimization Example: CRISPE Optimization Example: RASCEF Optimization Example:

​

Related Documentation

• To create custom prompt templates programmatically via API, see Creating Custom Prompt Templates via API.
• To view, manage, and use custom prompt templates, see Managing Custom Prompt Templates.