Behind the Quill: Character Portraits
- Erin H
- Aug 24
- 4 min read
Another thing I’ve been working on—though not as much recently—is character and estate portraits which I've already been posting in the gallery.
Before I started the Behind the Quill series, I spent several weeks focusing on them. Previously, I’d spent a long time working on character references. I searched the internet, using face-merging programs to try to mesh people together into new faces. It usually resulted in very vague, alien-looking images. I also tried finding real pictures of real people or AI-generated people that I could use as general references. For more important characters, I had up to 20 images each of what they could potentially look like. It was a bit insane.
However, with the use of AI, I’ve been creating actual character portraits. I know a lot of people are against using AI to create images, but I’m not talking about illustrations for publication. If you go to the book page on my blog, you’ll see that I’ve used images created by the Wix AI “Create an Image” tool. These are just visual references—not final book covers or commercial artwork—just something to give a general impression and decorate the blog.
This an AI-generated portrait of Jane Austen and her siblings. These portraits were made by me, using the AI tools. Many people have tried to generate portraits of Jane Austen and Cassandra Austen based on her original sketches and descriptions, and the ones I made are similarly inspired. Meanwhile, there were reference images for most of her brothers and her parents, allowing me to make more realistic family portraits for them all.
Using the same style and process, I created character portraits by starting with all those reference images I had gathered. Since I had already merged my characters into real historical families, I first searched Google for any portraits of their family members. In some cases, especially with female ancestors, I couldn’t find anything. It seems women weren’t often painted unless their husbands really, really loved them—or maybe the portraits were destroyed or are kept in private vaults. Who knows.
If I couldn’t find a portrait of a grandmother, I would look for her father or her brother instead. If I found something, I’d either merge the father and brother or just make a female version of the father and use that as a placeholder. In this way, I was able to get a rough idea of what the grandparents might have looked like.
Once I had reference portraits of the grandparents, I’d merge them to create a hypothetical son or daughter. If a family only had one son or daughter, I’d just use that result. But in larger families, I’d then merge that generated son with additional references to create multiple brothers, giving them shared features but slight differences. I’d do the same for the daughters.
For background characters, I usually just created a “first daughter,” “second daughter,” “third daughter,” etc., changing the dress color and age to distinguish them. For major characters, I used these generated children and merged them with the AI references I already had to bring out more unique and realistic features.
While some characters don’t look quite how I originally imagined them, seeing their portraits helped me reimagine them. You may notice some resemblance among certain characters, especially if I used only one real-life reference. In most cases, though, I used at least two or three, especially for final portraits.
It must be said that, I’m not using AI to replace illustrators or take anyone’s job. Most people don't have the money to commission portraits for over 100 characters. I’m not a billionaire. If I were, I’d be publishing books a lot faster because I’d have a lot more free time. These portraits are for personal use, or for readers to reference while reading. That’s all.
That said, don’t think it’s easy to make good AI portraits. Sometimes AI has huge glitches. Sometimes you get the perfect portrait, but if you don’t download it fast enough, it disappears or gets corrupted. Sometimes it suddenly goes grainy. Sometimes, even if you specify historical hair and clothing styles, it won’t follow instructions—especially when real people are used as references.
For example, you may notice that Mary Elliot from Persuasion has short-cropped hair in her portrait. That’s not accurate for the period. But one of the reference photos I used had short-cropped hair, and I couldn’t get the AI to create a long-haired version without drastically changing her features. So sometimes you just have to choose the lesser evil.
Disclaimer on Character Portraits
The character portraits shared on this blog were created using a combination of historical paintings, publicly available photographs of real people (living and deceased), and AI-generated composites. The images were developed through a process of facial reference merging and AI generation, with the final selections made by me. However, the generation process itself was handled by AI (ChatGPT's image tools), which produced multiple drafts from which I selected the most accurate or expressive representation for each character.
These portraits are intended for non-commercial, illustrative purposes only, to help readers visualize the characters in my writing. They are not to be used for commercial reproduction, redistribution, or any purpose beyond personal or academic reference.
Any resemblance to actual persons—living, deceased, or historical—is purely coincidental and not intended to misrepresent, parody, shame, impersonate, or commercially exploit those individuals or their likenesses. No portrait has been sexualized, altered for malicious intent, or used in any way that would bring harm or discomfort to the individual(s) who may have partially inspired the image.
All images remain subject to OpenAI's Content and Use Policies. As the AI-generated images were created using ChatGPT's image capabilities at my direction, the resulting images may be considered co-generated content. According to OpenAI’s current guidelines (as of July 2025), users generally retain usage rights to content they generate, including images, as long as they comply with OpenAI's terms (in this case © Eireanne Michaels (Erin M Harris) 2024. All rights reserved.) However, because these portraits are based in part on historical and real-world references, I ask that others do not reuse or repost them outside this blog without permission.
If you have concerns about a specific portrait or its resemblance to a real person, please contact me so I can address it.
Comments