The Secret to Perfect Prompts (Without Prompt Engineering)

Key Concepts

Prompt enhancers, prompt engineering, text-to-image, text-to-video, LLMs (Large Language Models), Midjourney, custom GPTs, projects, prompt structure, few-shot prompting, chain-of-thought reasoning, role-based prompting, structured output formatting, creative brainstorming, scripting short-form content.

Text-to-Image Prompt Enhancer

Goal and Structure

The goal is to input a basic prompt and receive a more complete, well-formatted prompt suitable for image generators like Midjourney. The enhancer is broken down into three parts:

1. Core Purpose: Defines the role of the LLM as a prompt enhancer, emphasizing its purpose and basic tips. It instructs the LLM to only respond with the enhanced prompt for easy copy-pasting. It also instructs the LLM to use best practices for AI image generation.
2. Prompt Structure: Provides a template for the enhanced prompt, including:
   • General overview of the scene (recognizable archetypes and keywords)
   • Subject details
   • Setting details
   • Style details
3. Prompt Examples: Includes 12 diverse prompts demonstrating proper structure and formatting (few-shot prompting).

Key Tips and Examples

• Conciseness: Instead of "Show me a picture of lots of blooming California poppies, make them bright vibrant orange, and draw them in an illustrated style with colored pencils," use "Colored pencil illustration of bright orange California poppies."
• Focus on what you want, not what you don't want.
• Balance directness and detail.

Usage

1. Direct Prompting: Copy and paste the entire prompt enhancer instructions into a chat with an LLM (e.g., ChatGPT, Claude, Gemini). Then, provide your initial prompt.
   • Example: Initial prompt: "A wizard casting a spell."
   • Enhanced prompt: "A bearded wizard in a flowing deep blue robe stands at top a windswept cliff, one hand raised as he casts a glowing arcadan spell into the stormy sky. Magical runes swirl around his fingertips crackling with blue and violet energy. His eyes glow with power beneath a wide-brimmed hat and his staff pulses with light. Dark clouds churn above, lightning flashes in the distance. Dramatic fantasy lighting, high-detail digital painting, cinematic composition inspired by classical fantasy art."
2. Project/Custom GPT:
   • Create a new project or custom GPT in ChatGPT or a similar platform.
   • Add the core purpose as instructions.
   • Upload a file containing the prompting tips, tricks, and example prompts as a knowledge base.
   • Test the enhancer with basic prompts.

Benefits

• Improved image generation results (more controlled and specific scenes).
• Better organization of prompts (especially with projects).
• Easy access to the enhancer (pinned to the sidebar).

Text-to-Video Prompt Enhancer

Similarities to Text-to-Image

The process is the same as the text-to-image enhancer, but the instructions and prompt structure are tailored for video generation.

Key Differences

• Prompt Structure: Focuses on camera movement and scene description. Primarily draws from Runway's documentation on prompting.
• Multiple Prompt Options: The enhancer is instructed to respond with three enhanced prompt options, allowing the user to pick the best one or combine elements from each.

Usage

The usage is the same as the text-to-image enhancer, either through direct prompting, a project, or a custom GPT.

LLM Prompt Enhancer (General Purpose)

Goal

To create an LLM that can analyze a user's request and refine it into a more effective prompt, adapting its approach based on the prompt type.

Steps

1. Determine the Best Refinement Approach: Analyze the user's request and determine the most effective way to improve clarity, specificity, and structure. Choose the appropriate approach based on the prompt type (e.g., summarizing an article vs. creative brainstorming).
2. Apply Industry Standard Prompt Engineering Techniques (Only When Needed): Use techniques like few-shot prompting, chain-of-thought reasoning, self-consistency prompting, role-based prompting, and structured output formatting only when appropriate.
3. Deliver the Final Refined Prompt:
   • If the prompt is already clear, refine it and provide the optimized version.
   • If additional information is needed, ask only the most necessary questions before refining the prompt.
   • If the request is overly broad but usable, provide a refined version and mention that more details could improve accuracy.

Examples

• Initial Prompt: "Explain Koji economics."
• Refined Prompt: (Generated by the LLM) A more detailed and structured prompt explaining Koji economics.
• Initial Prompt: "What are the most common programming languages?"
• Refined Prompt: (Generated by the LLM) A prompt asking for a markdown table listing the most common programming languages.
• Initial Prompt: "Give me app ideas I can vibe code."
• Refined Prompt: (Generated by the LLM) "Give me a list of fun and creative app ideas that are simple enough to vibe code, quick to prototype, minimal setup, and fun to build solo. Focus on small scope, original or quirky concepts, and room for a playful UI or interactions. Present that in a bulleted list with a short one to two sentence description for each."

LLM Prompt Enhancer (Specialized - Scripting Short-Form Content)

Goal

To create a specialized LLM that assists with scripting short-form content by providing step-by-step guidance and brainstorming ideas.

Process

The process is structured as a back-and-forth conversation with the LLM, with each step building upon the previous one:

1. Initial Input: The user pastes an article, tweet, or concept.
2. Hook Ideas: The LLM provides 16 different categories of hook ideas (e.g., shocking statistic, intriguing question, bold statement).
3. Visual Hook Ideas: The LLM provides ideas for visual hooks to grab attention.
4. Key Bullet Points for the Script: The LLM generates key points to include in the script.
5. Ideas for Curiosity Gaps, Unexpected Twists, Escalating Reveals: The LLM provides a list of ideas to enhance engagement.
6. Rough Script Outline: The LLM generates a rough script outline based on the previous steps.
7. Closing Statement Ideas: The LLM provides ideas for closing statements (e.g., call to action, loop creation, thought-provoking question, cliffhanger).
8. Final Review: The user reviews the generated content.

Example

The example used in the video is based on the topic of "the secret to endless prompts." The LLM generates various hook ideas, visual hook ideas, key points, and other elements to help create a compelling short-form video script.

Conclusion

The video demonstrates how to create custom prompt enhancers for text-to-image, text-to-video, and general LLM usage. By tailoring the instructions and providing examples, users can significantly improve the quality and relevance of the prompts generated by LLMs, saving time and effort in the process. The video emphasizes the importance of prompt engineering and provides resources for further learning. The key takeaway is that customizing LLMs for specific use cases can lead to more efficient and effective results.