When Open AI announced that they were going to incorporate Dall-E 3 into ChatGPT-4 back in late September, I was in the process of learning about Midjourney. I had seen several people use Midjourney to create images for Graphic Novels and I saw how I could do the same. The thought of having ChatGPT-4 and a solid image generator like Dall-E 3 had me rethink things and roll with Open AI.
I didn't get access until the second week in October, but once I did I was in love. Like Midjourney, I was receiving four images at a time. You could have ChatGPT-4 help you create the prompt or provide it with the exact prompt you wanted it to send to Dall-E3. I knew within the first week that it was all I would need to start adding images to my written work.
Within two weeks I was cranking out all kinds of images and I felt supercharged just like when I first started using ChatGPT-4 in Spring. However, by the end of October, Open AI started messing around with things. The next thing I knew I was no longer able to use seeds to try and maintain consistency and they had reduced the image output from four images to one. They had kneecapped me literally as I was starting work on creating images for Michaelmas.
I felt betrayed and let them know about it. I was still able to create some images but now it was a trickle and the capped messages meant that once I started being able to create the images I needed I had to wait a few hours before I could start creating more images.
Open AI then decided to do another update to their models before making things right with Dall-E 3. We got the GPTs, which are really cool but even to this day you can only get one image at a time with the main model. You can use the Dall-E 3 GPT that they released which provides you two images at a time, but now this new ChatGPT-4 Turbo model is not quite as intuitive as the older model. It often makes you explain things more than once to get "Turbo" to do what in the past only took one try. And the message cap is even lower now because everyone thinks they are going to create a money-making GPT. So much so that Open AI had to cut off new registrations because they were stressing the servers.
Yet, I kept at it with Open AI even though I was only receiving one or two images at a time and all the other frustrations. Then they fired Sam and then rehired him during a crazy weekend which made me question my decision to roll with Open AI to create my artwork. Thank goodness for Microsoft Copilot, which over the past few weeks has become a reliable fallback option to the unstable chaotic mess going on with Open AI. This is a bit ironic considering Copilot is ChatGPT-4 Turbo with Dall-E 3 and provides four images at a time just like Open AI did when they first released that model in October.
Two things suck about creating with Copilot: You can only get the images in square format, and I need them in portrait format. Also, you cannot receive Seeds or Gen Ids for the images you create. This makes it more difficult to create consistent characters.
Why did I mention all of that, because all of that crap has slowed down my work on creating images to a crawl over the past month and a half. Had Open AI not messed around with a good thing I would be further ahead of where I am now and may have been able to release the illustrated version of Michaelmas before the new year and possibly make some money during the holiday season.
Do I wish I had stuck with Midjourney or switched to a different AI Generator like Leonardo AI instead of rolling with Open AI through the chaos of the past few months? It's not like people are begging for new material from me like I'm George R. R. Martin, but I would like to move on to working on my first Graphic Novel once I have released the Illustrated version of Michaelmas. Will I stick with Dall-E 3 as my main image generator beyond Michaelmas? I am weighing up my options, and there are a number of them.
I am grateful for what I have learned this year about Gen AI. Though I have dropped the ball on my writing to do so. My reasoning was that once I incorporate AI into my workflow I can write at least twice as much as I had been able to before my introduction to Gen AI. Hopefully, that will be true but I need to get back into the flow.
Over the past few years, I have been setting these deadlines to achieve certain goals and I have been falling short far too often. The three TV Series I had been working on to pitch are a perfect example. That kept getting dragged out until this Spring when the WGA went on strike and I decided to focus on Gen AI to transform those stories into graphic novels. Maybe the goals I am setting are too unrealistic. I've always been self-motivated to create so I can't complain about not having fans hounding me for my next story. Who knows what that pressure would do to me anyway.
While it kills me to not have all of the images ready to go before Christmas, I do have a load of other images that I want to start sharing. Most of September and October were spent learning about Midjourney and Dall-E 3 by creating all kinds of different images. I wanted to understand how these images were created and learn the different styles I could use. While I settled on a more historical look for the images I also wanted to experiment and made an effort to avoid focusing on Michaelmas images until November. By that point, Dall-E 3 was only doing one image at a time and I was fuming mad over that. And then literally the weekend before I was to start on Michaelmas images Dall-E 3 decided to again change a major part of how users communicate with it, removing the ability to use parameters like seeds, which threw me into a tizzy. So from the very moment I wanted to start creating Michaelmas images with Dall-E 3 all my research went out the window.
Eventually, they got their act together after I lost a week plus because of the Gen ID implementation nonsense, where you could only use Gen IDs instead of Seeds. I love Open AI but they are frustrating as hell sometimes. We are all trying to learn how to use these awesome tools they have created and then they go and change it every other week in ways that are beyond frustrating.
That said, it's amazing how lifelike images have become. Dall-E 3 is okay with lifelike close-ups whereas Midjourney, Adobe Firefly 2, Leonardo, and others are actually a bit better. Since the images I am creating are meant to look like they were taken back in the day I am more concerned with prompt adherence than lifelike images. Dall-E 3 excels at a lot of things that others don't come close to.
The image at the top of this page was one of the first Michaelmas-themed images I created. At the time, colorful paintings similar to those from that period really appealed to me. For the first few weeks, I was creating similar images with loads of color that looked like they could have been hanging in a museum.
If you want to create amazing images using AI, I would strongly advise following people on Twitter/X who constantly upload images and the Prompts they use to create them. There is so much to learn there. I have gone through and bookmarked thousands of tweets and made documents based on those that I really like. Much of the work I have done over the past few months was inspired by what others were doing. And there are some amazing AI Artists out there and you can learn a hell of a lot by simply following them and what they do.
