I wrote the prompt exactly as I saw it online, but the image that came out was completely different…
Doing it my own way didn’t work, and following the tutorial articles didn’t result in my ideal image…
When using image generation AI (like ChatGPT, Midjourney, or SeaArt), you often run into this problem where ‘following the know-how doesn’t work’.
I was exactly one of those people.
At first, I tried trial and error on my own, but I crashed and burned.
Next, I scoured the internet for ‘god-tier prompt collections’ and tutorial blogs, but even after copying them exactly, I never got satisfactory results.
When I was stuck, thinking, ‘I don’t even know what I’m doing wrong,’ I suddenly changed my perspective.
‘It’s not working because I’m trying to come up with the prompts myself. Why not have ChatGPT create a rough image first, and then have ChatGPT itself explain the ‘composition (prompt)’ of that image?’
This ‘reverse engineering (asking for an explanation)’ approach was a hit!
By having it generate a rough image once and then asking ChatGPT, ‘List the prompts (English words) for this image’s hairstyle, clothing, background, and lighting part by part,’ I was able to see at a glance exactly what I needed to rewrite to get my ideal image.
It takes a little time, but I became able to create images exactly as I intended, and by saving the English words it listed in my smartphone’s notepad, future image generation became surprisingly easy!
In this article, I will reveal the ‘reverse engineering technique to have AI dissect image composition’ that I arrived at, as well as the ‘part-by-part god-tier English word notepad’ that I actually save and find useful!
I. Why does ‘copying prompts from the internet’ not work?
The reason why copying and pasting prompts found on the internet doesn’t result in the image you intended is simple.
In other words, even if you copy the finished product (the prompt), you cannot apply it because you don’t understand the ‘structure’.
That is where the method of ‘making ChatGPT your image dissection assistant’ comes in handy.
ð¡ 3 Steps to ‘Reverse Image Generation’ That Won’t Fail
-
Step 1: First, have it create an image with a rough instruction
-
(e.g., ‘Draw an illustration of a cute Japanese girl in a cafe’)
-
-
Step 2: Ask for the ‘composition prompt’ for the generated image
-
(Ask it to ‘List the hairstyle, clothing, background, and art style of this image as English words for each part’)
-
-
Step 3: Look at the listed parts and modify only the words you want to change
-
(e.g., change ‘white shirt’ in the clothing section to ‘pink hoodie’ and regenerate)
-
You don’t need to think of English from scratch, nor do you need to decipher prompts from the internet.
You just need to have the AI explain the picture it drew itself!
The content beyond this point is in the paid area.
I will reveal the ‘Image Anatomy Prompt’ that you can use as-is and the ‘English Word Part Dictionary by Category’ that you can save to your smartphone and use for a lifetime!
