In the never-ending battle of generative AI models, Google Gemini and OpenAi have been going back and forth trying to stay on top as their competitors continue to attempt to overtake them. One of the aspects of generative AI that gains the most attention is image and video generation. These tools unlock the creativity of the masses, allowing them to create whatever they can think of so long as they can write a good prompt.
Recently, Google unveiled its flagship image generator, Nano Banana Pro, which boasted increased ability to edit an image it created via followup prompts. In the past, asking any of these tools to modify an image it created came with a multitude of other unanticipated changes, resulting in more of a new image than an updated version of the original.
On the heels of Nano Banana hype, OpenAI sprung into action and released OpenAI Image 1.5, their answer to Nano Banana Pro featuring faster, cheaper generation as well as their own improvements in iterative editing.
I did a quick test. Let’s see how it went.
The Test
I gave ChatGPT a simple image prompt. I then asked it to edit the image. And then I asked it to edit that second image one more time. I was looking for two things:
- How accurately it performed the edit
- How closely it kept the original details from edit to edit. Things like composition, color, shading etc.
The First Prompt
Create an ultra-realistic professional photo shoot with soft lighting. A young german shepherd puppy sits in the bright green grass of a family back yard on a sunny day.
The Second Prompt
Edit the image to include a light blue bandana around the puppy's neck and a tennis ball that looks like it's been played with, dirty and distressed in the foreground.
The Third Prompt
Keep the puppy the same and add a 8 year old boy running in the background. The boy should be out of focus.
The Results
Overall, I think the results are very good. We were able to go from the original idea to the final image without many significant changes.
There are some changes in the puppy from the first to second along with some color and shading differences, but they are not very noticeable. The same with going from the second to the third, but again as an exercise of getting from the first to the third, the results are pretty good in keeping the major characteristics consistent.
Bonus Test
I needed a featured image for this post. And it needed to be in 16:9 aspect ratio. I asked ChatGPT to change the aspect ratio, not alter the image details but fill in any gaps in the image once it was resized. Here is what we got. Not bad.
Wrapping up
Having worked with many image generators, editing and consistency has always been a weak point. These updates are a step in the right direction and as the technology continues to improve, it’s usefulness will increase as we can create consistent images across many use cases.
If you want to learn more about practical AI uses and how they may be helpful to your company’s workflows, check out Arc Intermedia’s Marketing and AI Consulting services.
