With the integration of Dall-E into ChatGPT, you can ask ChatGPT to modify an image based on your feedback. This allows for the creation of a series of consistent images. Imagine wanting to retain the subject in a series of images while making changes to other elements like the background. This can be easily done.
In this post, I’m going to show you three ways to do it.
When There’s No Consistency
In the absence of consistency, every time you give ChatGPT a prompt, such as “tourist taking a selfie with the Eiffel Tower in the background“, it produces an image of a completely different person, as shown in the following examples:
Why Aim for Consistency in Image Creation?
Imagine wanting to generate a series of AI images where the main subject remains consistent, while other elements vary. This approach is ideal for thematic collections, storyboards, or any project where continuity is crucial. The ability to make targeted adjustments to your generated images ensures the preservation of the core theme, while also allowing for creative variations.
Here’s an example where I have ChatGPT strive to keep the subject (the tourist taking a selfie) the same, while changing the background to include different attractions.
3 Ways to Achieve Consistency in Dall-E
Here, I’ll demonstrate with examples three easy ways to achieve consistency in my subject with prompts.
The goal is to maintain the subject while altering the background to feature different tourist attractions.
1. Using Seed
Think of the “Seed” in Dall-E as a unique starting point for the AI to begin its image creation journey. It’s similar to choosing the first note in a song or the first brush stroke in a painting.
When you use the same Seed with the same prompt, Dall-E will consistently produce the same image, or ones that are strikingly similar. This is super handy if you’re looking to tweak an image but want to keep some elements the same.
After you’ve got your first image, just ask ChatGPT, “What’s the seed number for image 1?“
Using the seed number, you can create follow-up prompts like this:
Using seed: [seed_number], the same tourist taking a selfie with [attraction] as background
After a couple of prompts, here are the outputs I received:
The subject remains fairly consistent, though sometimes the AI gets confused about whether to position the phone facing front or back.
2. Prompting “the same…”
Another method is to use simple, natural language instructions, asking ChatGPT to keep the subject the same while only changing the background. After generating the first image, to ensure consistency, I prompted:
the same tourist taking a selfie with [attraction] as background
And here are all the results:
With this method, it seems that while maintaining consistency on the subject, it also preserves consistency in how the phone should be placed, whether front or back.
3. Prompting “change…”
The final method I tried, which I believe offers the best consistency, involves directly instructing ChatGPT on what to retain and what to change. The prompt I used is simply:
same tourist, change background to [attraction]
Here’s what the images look like:
This approach appears to effectively get the job done.
If you’re looking to achieve consistency in image creation using Dall-E within ChatGPT, I hope this post has provided you with some helpful insights. Although these are just three methods, I believe that with a bit of experimentation with prompts, you can effectively guide ChatGPT to produce more consistent images.