Creating consistent AI-generated character designs using text-to-image generation AI generated text to image Character Design - Fred & Eric

AI generated "handmade" character

Over the last few weeks, we've been experimenting with the latest AI tools to see how they might be used specifically for character design and animation. (Yes we are perhaps being naively optimistic the tools will remain tools... and not take over completely!)

From AI illustrations, animations, post-production tools, writing, sound design, music and voice-overs; the technology is so fast-moving, this article will soon become a brief snapshot in time for the AI revolution. But we felt the era of misshapen hands and glitchy eyeballs was worth capturing before AI imagery becomes indistinguishable from real life. We're keen to explore the pros and cons of this technology, which is already becoming integrated into our design toolkit.

To do this, we've set ourselves the task of designing an AI-generated 3D "handmade" F&E style character, as a way to explore the technology.

Text to Image AI Character Design

AI image generation is addictive. You type in a description and what you get (for better or worse) is a complete surprise. This mini reveal keeps you generating more and more; each time refining the "prompt", by adding new keywords, and playing with style, texture and quality of finish. But what happens if you wanted to be more specific? Our character design project was about to begin. Here are a few of our opening gambits generated using Runway Gen-2 (March 2023).

Super creepy text to image AI character designs.
AI "cute" handmade character designs. (Generated April 2023)

"Cute" they are not (spot the cuddly creature who has two tiny arms for a bottom lip)... So far the results were hilarious/terrifying but not a patch on our human creators! Perhaps we'd given Chucky a run for his money though...

Nightmare-inducing results aside, the image-generation addiction had well and truly kicked in. We were on an image generation character design blind date, and each time Cilla revealed who was behind the door, we kept wanting more! And with options to create variations on preferred images, enhance or extend the backgrounds and so much more, it proved very hard to stay focused. After a few time-consuming tangents, on we went with with our open brief of generating a decent "handmade" AI-generated character design. This time we turned to Shutterstock AI. It came up with a few interesting designs:

Handmade Character designs generated using Shutterstock AI
Handmade Character designs generated using Shutterstock AI (Generated April 2023)

But for every one of these, we had to look at approximately 30 of these. Maybe you like them, but they are certainly not our style.

A collection of very creepy AI generated character designs.
"handmade" and "cute" character designs courtesy of Shutterstock AI (Generated April 2023)

This is when we moved over to Midjourney to generate our imagery. With just a few clicks and some creative commands, we met these guys. Now things were getting interesting / scary.

AI generated
"Handmade" character designs created using text-to-image in Midjourney (Generated April 2023)

Creating consistent character designs using AI

Vexxer - and AI generated 3D character made from wool.

We immediately loved this creation. Cute and cool; handmade but not twee. Just what we were hoping for. Chat GPT named them Vexxer, stating, "It's a short, punchy name with an edgy sound that could work well for a character with a mischievous or rebellious personality"...

Now our character had a name, task two was to see what we could do with this guy.

Creating consistent character designs

Using Vexxer's own unique seed number along with a fairly specific 'prompt', we started to expand Vexxer's repertoire. We always like to start with a few of our fails:

A set of creepy images which explore AI image generation gone wrong.
Character design fails (we'll not even get into the lack of diversity...)

Another tutorial video under our belts, and all of sudden, Vexxer's family had been brought into existence:

A troup of handmade character designs made from pastel coloured woollen tectures.
Consistent character designs that illustrated the image learning process working.

We'd undoubtedly created some cousins for the mighty Vexxer, but in terms of reproducing the exact design at multiple angles or in alternative settings, we'd not managed to succeed.


The potential for doing this though, is getting easier to use by the day. We're sure it won't be long before we can upload our AI character Vexxer, and create a complete 360 3D render at the click of a button. A few more clicks Vexxer could be fully rigged and have a whole range of pre-animated movements at the disposal of a creative text prompter.

Is AI character design the way forward?

We felt happy our newly created character family would be well at home in with F&E (with a bit of finessing). The design style felt unique and "handmade" in feel. It's definitely a character we'd have loved to have come up with... But had we come up with it? Who really was the creative behind this design? Did our creativity help to generate a character we were happy with, or were we lucky? We didn't input any original designs into this image generation. The process feels a little bit like cheating.

Plus, we haven't even touched on copyright, ethics, usage and environmental ramifications associated with generating imagery in this way. These consequences are as complicated and fast-moving as the technology itself.

What's next for our AI character design journey?

After all the intensity of staring at a screen (and the addictive nature of image generation) we've vowed to do some serious carbon off-setting for all of the processing power Vexxer's inception demanded. We're returning to the (real-life) knitting needles and marker pens for our next character project!

However, our journey has certainly been enlightening. As an open brief, we love the design of this character. Perhaps though, if we had a more specific brief and a clear creative vision, generating imagery in this way could be a total time-waster. As always with new technology, we will find a way to use these tools as part of a wider creative process, without feeling we are 'cheating'. Trusting our own creative intuition and integrity to produce new and exciting work.

AI image generation (and animation) is only going to get more user-friendly and integrated into our working practice and lives. Better to understand it, than to be left behind. We've vowed to be open to the new technology, whilst staying true to F&E's handmade, craft roots. Whether you like or loathe AI, it's here to stay.

It’s 2023. We're just getting started.