A Tale of Four Platforms
During our last ActiveRain Zoom meeting, the topic was AI image generation. I've had a lot of fun with this over the last few months, but lately it's turned a little frustrating.
As many of you know, I’ve been using AI tools to create portraits of U.S. Presidents for my series. It was going great until recent policy updates changed what’s allowed. Suddenly, the ability to generate images of famous people, including presidents, became restricted on several platforms.
That shift inspired this week’s assignment: a comparison of three AI image generation platforms, plus a bonus one I explored out of curiosity. My lineup included ChatGPT, NightCafe, Meta.ai, and Artistly.
I used the following prompt for each image:
A watercolor portrait in the image and likeness of President Dwight D. Eisenhower. United States flag in the background.
ChatGPT
This is where my experiment began.
I have found ChatGPT’s built-in image generator to be intuitive, conversational, and surprisingly accurate, until the recent policy shift. After the trouble I had earlier in the week, I was surprised by how accurate this experiment turned out.
Perhaps the simpler I make the prompt, the better. This contradicts many things I've read that the more details, the better. Apparently, that's not true with people. However, this is just one example. Who knows what will happen when I do the next one.
So, for now, I'm going to keep it simple when describing people. 
NightCafe
NightCafe leans heavily toward artistic styles, with many options beyond the actual prompt.
I've had some nice results with various images, but not real people. It is not conversational or interactive like ChatGPT. ChatGPT remembers what I did earlier, while NightCafe does not. You have to keep tweaking the prompts and trying over and over, which eats up credits... and time!
Speaking of credits, I use the free version of NightCafe. I log in every day to pick up my daily credits, and I sometimes participate in their daily challenges to earn additional credits. I have accumulated several thousand credits, so I'm not really worried about using them all up.
I know there is a paid version of NightCafe, but I haven't been tempted to go there... yet. I think Margaret Rome Baltimore 410-530-2400 is better versed on this particular platform.
Overall, the results were nice but less realistic. I did find myself appreciating the “artistic interpretation” even if it didn’t look like the real person. This would work great if you want to depict a person, but not one specific person.
One thing I really like about NightCafe is when you go back to look at previous images you've created, the whole prompt is there for you to see. This helps if you want to recreate something similar or maybe change it a little bit without trying to remember how you did it.
Meta.ai
Meta.ai had an entirely different vibe. It produced four images quickly and leaned toward a stylized realism, though the likeness to how Eisenhower actually looked wasn't as accurate as ChatGPT. Accuracy varied, but ease of use was excellent. And, I like having four options to choose from.
Artistly (Bonus)
Our assignment was to cover three different platforms, but being the overachiever that I am, I tried Artistly as a bonus.
This is a platform I haven't used hardly at all. There are many options I'm not familiar with, so it's hard to rate. Given my limited skill with this platform, the results weren't something I'd use in my presidential series. If I were creating a person only based on gender, age, setting, era, etc it would probably be fine. It's just not what I need for my particular project.
So, the winner is...
Each platform interpreted the exact same prompt differently. As frustrated as I've been with ChatGPT the last week for this particular presidential series, today it gave me the best results for what I want. Of all the platforms, ChatGPT takes the longest to generate an image.
Oh, and I tried to see what the Bing image generator would give me, and it blocked the prompt completely. I know Bing is a platform that Kat Palmiotti likes. I will experiment more with the other platforms for other projects that don't involve a real person.
All of this is a great reminder that AI art reflects algorithms as much as imagination. Policies will continue to evolve, but exploring these tools has given me a basic understanding of their current strengths and limitations.
Thanks for the Challenge, Debe Maxwell, CRS Assignment complete.
Comments(18)