I’ve checked Reedsy, Seventh Sanctum und Reddit – it’s nice to have, but nowadays, AI can give you very specific writing prompts within seconds. I think that this is now the most effective way to start creating story ideas.
Checking out Notion will cost me more time. So far I’ve used OneNote. But it seems like the basic Notion plan is free. If it offers everything that I need, this could be another reason to abandon Office365.
The tips on categorizing ideas is a good one to explore. I simply wrote everything thing down and have a giant bullet point list now which is hard to navigate.
Once I’ve got a better system setup, I’ll write a more conclusive post about creating an idea bank.
You can follow my Grok Diary in real time here: @michael_brig
My first attempt with ChatGPT (Version 5) is finished — The Last Superhero Part 3.
Right now, I’d say the result is maybe 20% of what I would like it to be. I plan to create Part 4 in the next couple of weeks and try to improve the results, as there is still some room for experimentation.
For now, here are the biggest problems I encountered.
Just like Midjourney, ChatGPT has a very strict language filter for image prompts. For text generation, you can tell ChatGPT that you’re working in a fictional setting. That allows you to describe certain acts of violence or crime to some degree.
With image generation, however, this isn’t possible at all. Even hinting at violence in a comic-book context can trigger the filter.
For example, I had problems generating an image where a character gets water splashed onto his face. That alone triggered the system.
The same happens with facial expressions. Pain alone might work, but pain combined with bruises often gets flagged — even without describing the action that caused them.
Don’t even try using the free version.
You might only get two or three images every couple of hours. For my 31-page comic, I generated more than 120 images.
Even the paid version has timeouts. After roughly every 20 images, ChatGPT asked me to wait a couple of hours before I could continue generating more.
Midjourney handles this much better — especially considering that the prices are somewhat comparable.
You can clearly see how the comic switches between different art styles. I tried to anchor the prompts around a specific comic-book artist, but every few images the style drifted again.
I’ve seen this with Midjourney as well. When prompts become too long, parts of them seem to fade away and become irrelevant. The AI then simply ignores those sections.
Clothing and the general appearance are mostly fine, but the face of my protagonist drifted quite a lot.
Character consistency remains one of the biggest issues, especially if you attempt to create something larger like a 160-page comic.
Facial details are very difficult to control. My character’s beard looks slightly different in almost every image, and the hairstyle of the female doctor changes frequently as well.
This is similar to the character consistency problem. The more detailed the environment, the harder it becomes to keep it consistent across multiple images.
It helps to generate all images within the same chat, but once the conversation reaches around 20 prompts, things start to slow down. The chat becomes sluggish and sometimes even glitches.
Overall, Midjourney still offers a better user experience. It’s easier to fine-tune prompts, results arrive faster, and the whole process feels more controlled.
There are quite a few issues. I think some of them can be improved with better prompting and a couple of workarounds.
For now, I would still recommend Midjourney for AI comics. That said, with a few adjustments I might be able to get better results with ChatGPT when creating Part 4 of The Last Superhero.
I’ve started making AI comics again. Last year was a good beginning, but I didn’t feel the models had evolved enough to justify continuing the project on a monthly basis. That may have changed in 2026.
The quality still fluctuates — to put it politely.
But the idea behind The Last Superhero series is to use it for experimentation and testing anyway. So let’s continue.
For Part 3, I used ChatGPT for the first time. It has moved beyond DALL·E 3, so this comic was created using OpenAI’s GPT-5 image model.
There are plenty of issues: I’ve planend on creating a 4th part using ChatGPT for March where I try to get around the most servere issues. But I will also write a future post where I discuss all the problems and possible solutions.
For now, here is the complete comic short story…
For step 1 AI models have become super helpful. Just tell Grok/ChatGPT/Claude to give you some prompts, pick one and start. Even step 2 can be assisted with AI. Step 3 is then the job of the author: Pick your favorite hook and store it in your swipe file.
You can follow the project in real time here: @michael_brig