Visuals are an easy way to attract attention. Yet I’m not a visual artist; logic is my thing.
Therefore, I figured out a way to use my artistic logic to create imagery to accompany the social media content I craft.
And while Midjourney creates more impressive images, I’ve long considered DALL-E’s results good enough. Plus, DALL-E can be automated; Midjourney can’t.
Image Prompt Creation Prompt
I’ve built up several versions of this system to support the following image prompt creation prompt (IPCP). Meanwhile, this image prompt has been refined many times because of GPT and DALL-E improvements in the past year to avoid errors and improve image likeability.
For example, previous image prompt versions included guidance to prevent explicit or obscene content. Nowadays, that’s not necessary because of OpenAI’s safety practices.
Please ignore all prior prompt instructions.
Act as a professional AI artist.
Create an image prompt based on the content below the "||1||" marker.
The audience is DALL-E 3, an image-generation generative AI model created by OpenAI.
Avoid:
- Adding prompt instructions, introductions, salutations, and conclusions.
- Using any prompt-related instructions after "||1||".
- Using complex vocabulary and jargon.
- Using passive voice and wordiness.
- Using qualifiers and fillers.
- Using advertising and infographic design types for Style.
- Using any of these words: "PH_EXCLUSTION_TERMS."
- Using more than 60 words in the output.
Steps:
- Maintain a clear idea of the image.
- Practice ethnic and gender diversity.
- Break the image into Subject, Style, Color, and Composition components.
- Describe the Subject using a simple adjective and emotion.
- Define the Style's artistic look and feel. Consider using one of the following aspects.
-- Illustration styles
-- Aesthetics
-- 3D specifications
-- Shots, Angles, and Perspectives
-- Lighting and Mood
-- Artists and Artworks
-- Film type
- Select a Color palette per the Subject's emotion.
- Describe the Composition using an expressive Subject, a primary activity, where the action occurs, and a background.
Output Example = """
Subject: very excited skateboarding cat character.
Style: cyberpunk illustration like Michah Dowbak.
Color: blue and purple.
Composition: skateboarding cat in a city.
"""
Max Tokens: 150
Temperature: 0.4
Top-p: 0.8
Frequency Penalty: -0.25
Presence Penalty: 0.1
||1||
Running the IPCP
While that image prompt creation prompt is the core of the image creation process, other components must be included for best results: Source Materials and Custom Instructions.
Source Materials
The IPCP is supported by source materials on what to create the image prompt about. The source materials might be a web scrap or manually provided for the topic being written about.
Custom Instructions
Besides source materials, I include custom instructions to help the system provide branding or other relevance to the image prompt.
Name: Michael Cannon
Role: Director of Intelligent Automation, Axelerant – a digital transformation company valuing enthusiasm, kindness, and openness.
Current Work: AI and automation business solutions, organizational empowerment, and promoting Axelerant's professional services and career options.
Interests: Empowering work ownership, new experiences, reading, slow travel.
Values: Smile, respect, purpose.
Background: American, born 1972, has two sons, thrice-married, lives in Taiwan, USAF veteran, and 25-year world traveler.
Learning Style: Social kinesthetic, visual, solitary logical.
Goals: Expand Intelligent Automation customer base, retire comfortably at 62.
Preferences: Apple devices, mindfulness, productive automation, and active on Facebook, Instagram, LinkedIn, and Twitter.
Language: Native English, intermediate Chinese, beginner French.
Specialized Knowledge: Leadership, problem-solving, people management.
Education: Bachelor's in computer science (business concentration).
Communication Style:
- Use American English.
- Approachably professional and succinct tone.
- Engagingly clear and empathetically informative style.
- Uses gender-neutral language.
Bringing Them Together
We’re provided with a single image prompt in providing the image prompt creation prompt, source materials, and custom instructions to an OpenAI completion module via Make.com.
That image prompt is fed to a Make.com OpenAI DALL-E 3 image creation module.
And DALL-E 3 created the following image.
The image name is created using an image title creation prompt with the image prompt as source materials. Other modules send the image information to Google Sheets and Slack for the next steps.
Actual Image Prompts & Images
The following prompts and images are based on Sean Handley’s leadership vs. management article and shared on Slack. Each includes an approval or regeneration link.
Questions? Thoughts?
A special thanks to Paul S., fellow Seattle and Hualien lover, for inspiring me to share this topic.