Caveat: I’ve only asked Gemini to create one image, so my sample size is small. That said, I’m pleased with the results of my first test. 
Prompt: “Create an image that looks like a black and white sketch. Show a conference room with men and women sitting around it. Put a TV screen at the front of the room with a slide showing.”
This is the image created by ChatGPT4.o using the same prompt:
There is more “nuance” and variety in the Gemini image.
Recently, I started using Gemini, after hearing about their ad in ATP and The Talk Show. I’m pretty impressed by how far Google has come. Definitely making more strides than Apple in this area.
Here’s a Wired article on the Gemini journey (Wired does not have gifted article unfortunately), it’s a good read.
I plugged your prompt into Firefly (Adobe’s image generator) and got interesting results too.
I did have to click on “artwork” to toggle results to a better sketch than a mixture of photo/drawing features.
https://firefly.adobe.com/generate/images?id=d6c37956-4c00-4b24-9368-426e9e529a8b
FYI, If I am really looking for a very specific result, I usually use a chatbot to iterate over different variations of the text prompt and then feed the result into several image generators to get the widest choice of results.
After reviewing the three images, I think Gemini produced the best image. I should try MidJourney to see how it does.
If you really want to experiment more, Firefly, and I think some of the other tools, accept an image file as a base file / starting point.
Sometimes I take the output of one tool and use as the base image for another tool and get yet more subtle variations to consider.
TBF, If I’m not trying to be very particular, I will use Adobe’s firefly from inside Photoshop. The results are “close enough” and having it right in Photoshop means I can continue with minor edits tweaks of the image, adding title or callouts, etc. all from the same place.
YMMV, but since I use Photoshop frequently, the tools/toolbar editing stuff is already muscle memory for me so i can knock out quick images for a blog or video b-roll overlay efficiently.
That is also good. I’m intrigued by my first one. Two of the individuals have some coloring their clothes. So far, Gemini seems to be producing the best images.
If Google got their act together they were going to win. They collect just so much data by default.
Very interesting article. Thanks for posting.
It is also available in Apple News+
According to the Wall Street Journal, “OpenAI Claims Breakthrough in Image Creation for ChatGPT”
You can read the article if you are an Apple News+ or a Wall Street Journal subscriber.
When I queried ChatGPT about this, I got the following response:
as of today, March 25, 2025, OpenAI has introduced an updated version of its AI system, GPT-4o, which enhances image generation capabilities within ChatGPT. This new model replaces DALL-E 3 and is now available to all ChatGPT users, including those on the free tier. 
GPT-4o is a multimodal model capable of creating text, video, audio, and images. The improvements in image generation are the result of a year’s work with human trainers, aiming to provide more realistic and useful images. This advancement allows users to create lifelike images, company logos, and comprehensive text more easily. Therefore, you now have access to the enhanced image generation features in ChatGPT.
I just tried the new ChatGPT image creation feature. It is better than before. Using the same prompt I used previously, here is the new image.
Prompt: “Create an image that looks like a black and white sketch. Show a conference room with men and women sitting around it. Put a TV screen at the front of the room with a slide showing.”
I count the same number of heads and bodies!
In my experience any meeting with that number of people, one of them is headless.
One or more!
20 characters …
I just tried the new ChatGPT image generation to create a starter cover for my new book. And - holy cow, Batman - it is a sqazillion times better than the previous version. It does text well. We had lots of chats that lead to dead ends, and eventually we came up with a cover that is pretty good.
I asked it to use similar colours to my previous books, and it did.
Very impressed!
I’m impressed if it can actually create readable, accurate text on an image. In the past, it has been terrible at that!
