Every great character needs a backstory, a look, and a personality—enter ChatGPT (with DALL-E 3 image generation). I started by telling it I was creating a workplace safety course and needed a character to guide learners through common hazards in an engaging, relatable way. ChatGPT's first task was to conceptualize this character.

The initial ask was simple: “Create a character for my safety course.” But from there, things got interesting.

I wanted more than just a generic avatar. I wanted a character learners would remember. So, I started refining my prompts. I asked ChatGPT to give my character a personality, unique traits, and a physical look. Here’s where the prompts began to evolve.

Initial Prompt: Laying the Foundation

The first thing I did was define the basic concept for Maxwell Safe-T. This prompt was all about the essentials: I wanted a safety inspector, someone knowledgeable yet approachable, with a look that would translate well visually.

Prompt Example:

“Come up with a character for my animation about hazards in the workplace. Describe enough of their physical traits and appearance that you could write a GPT prompt to generate a unique creative image of this person."

Refining the Prompt for AI Image Generation

With Maxwell Safe-T’s character traits and physical features established, I wanted to take the next step—having ChatGPT write a specific text-based prompt that I could use in DALL-E 3 to generate his visual representation. This prompt needed to clearly convey Maxwell’s appearance, down to his safety vest and boots, in a way that DALL-E could interpret and reproduce consistently.

Prompt Example:

"Using the physical features and attire of Maxwell Safe-T, write a text-only prompt that I can use to ask you to generate an image of the character using DALL-E 3."

ChatGPT then produced a highly detailed, structured prompt that focused on the character’s defining traits and visual markers.

This step ensured that when I moved to DALL-E 3, Maxwell’s image would align perfectly with his backstory and personality. Now, I had the final prompt ready to generate his visual in DALL-E 3!

Streamlining the Prompt for Visual Consistency

As Maxwell’s character began to take shape, I needed to refine the visual prompt. Since the original prompt was quite long, I asked ChatGPT to make the description more concise, removing any unnecessary details and keeping only the essential physical traits and attire. This would help ensure that when I generated different poses in DALL-E 3, the results would remain consistent without overwhelming the AI.

Prompt Example:

"Using the prompt you created, remove all the expletives and write a shorter prompt that includes the physical traits and attire of the character only."

By simplifying the prompt, I was able to create a clear and consistent visual reference for Maxwell Safe-T, ensuring that no matter how many versions or poses I needed, DALL-E would generate a cohesive character every time.

Step 2: Using DALL-E 3 to Bring Maxwell to Life in T-Pose

With Maxwell Safe-T’s personality and look finalized, it was time to move on to DALL-E 3 to generate his visual design. My aim was to create an approachable yet professional character with a quirky charm—and, crucially, get him into a T-pose for easy rigging and animation later. This would allow me to ensure the body parts were well-aligned for future movements.

Here’s where I refined the prompt even further to include not just Maxwell’s traits but also the art style and the pose:

Prompt Example:

“Create an image of Maxwell Safe-T in a modern vector art style combined with flat design elements, in a precise and unmistakable T-pose, on a solid white background. He is a safety inspector in his late 30s with medium build and short curly salt-and-pepper hair. He has a round face, thick, square glasses, and a slightly crooked nose. Maxwell is dressed in a high-visibility safety vest over a light blue work shirt and khaki pants, paired with worn steel-toe safety boots. His hard hat is adorned with stickers from various workplaces, and he wears a digital watch. In a perfect T-pose, the character stands upright with legs slightly apart and arms extended completely horizontally to the sides, ensuring all body parts are visible and symmetrical. A neutral facial expression, facing directly towards the camera/viewer. Hands should be open with fingers slightly separated. The background is solid white, focusing entirely on the character.”

The Challenge: Getting the AI to generate a precise T-pose that works perfectly for rigging later. This part took a few tries because DALL-E didn’t always get the arms exactly where I needed them on the first go, but after a couple of prompt refinements, I got Maxwell’s T-pose to be just right.

Tip: When generating characters for animation or rigging, always ensure the prompt clearly describes a T-pose with all body parts visible. This makes it much easier when separating limbs and rigging for animation.

With Maxwell Safe-T’s design finalized, I needed to ensure the image had a high enough resolution to work with for animation. To do this, I used Upscayl, a tool that helped me upscale the image to a higher resolution before importing it into Photoshop. This gave me a sharper, more detailed image to work with when separating the character into individual parts for animation.

Cutting and Separating the Body Parts Once in Photoshop, the next step was to separate the body parts—primarily the arms, head, body, and various facial features. The more individual elements I could isolate, the more flexibility I would have when animating Maxwell later. For example, by separating the eyebrows and eyes, I could create more dynamic facial expressions. Here’s where the importance of layer naming really came into play.

Photoshop Layer Setup:

• +Max (the main character)

  • Head (Folder)

    • +Mouth (Folder)

    • Head

    • +Right Eye (Folder)

      • Right Blink

      • +Right Pupil

      • Right Eyeball

    • +Left Eye (Folder)

      • Left Blink

      • +Left Pupil

      • Left Eyeball

  • Body (Folder)

    • Body

    • +Right Arm

    • +Left Arm

The cutting process was critical. Using the lasso tool and pen tool, I carefully separated Maxwell's arms and head to create clean, isolated layers for rigging. But the real challenge came with the eyes. AI doesn’t generate high-resolution images, especially when creating full-body characters, which means the eyes often lacked detail, appearing distorted and unusable in Adobe Character Animator. The eyes simply weren’t clear enough for the software to recognize and animate properly.

Here’s how I solved it:

I asked AI to generate an image of Maxwell’s torso and head. This close-up provided a higher resolution, giving me eyes with more detail. Although the character wasn’t identical to the full-body version, the eyes were a close enough match. I used the pen tool again to cut out these more detailed eyes and replace the original ones in the full-body image. This gave me the resolution and clarity I needed for smooth, realistic eye movement during animation.

To ensure smooth speech animations for Maxwell Safe-T, I used the free mouth pack available from Adobe Character Animator. This pack includes a set of pre-built mouth shapes that correspond to different phonemes, such as "Ah," "Ee," and "Oh." Dropping it into my rigged character setup made syncing Maxwell's dialogue with his mouth movements much easier, ensuring fluid and realistic speech.

For those looking to implement this, you can refer to the Adobe Character Animator documentation for detailed instructions on how to use the mouth pack effectively in your own animations. It’s a quick and reliable way to add professional-level lip-sync to your characters.

Download the free Character Animator Mouths Pack: https://adobe.com/go/chresources

Step 4: Rigging Maxwell in Adobe Character Animator

Now comes the fun part—bringing Maxwell to life in Adobe Character Animator! I imported my Photoshop file (with those perfectly named layers) and started rigging the character so he could move, talk, and…well, do whatever safety inspectors do.

Tip: When rigging, pay attention to parent-child relationships. For example, the head should be a “child” of the torso, meaning wherever the torso moves, the head follows. This ensures smooth and realistic movement when Maxwell starts walking or gesturing.

The Challenge: Rigging those tiny details. Sometimes the smallest adjustments—like adding pins to knees or elbows—can make a huge difference in how natural your character moves.

Final Thoughts

Creating characters with AI, like Maxwell Safe-T, is a fascinating journey that mixes creativity with cutting-edge technology. It’s not always smooth sailing, but the results are worth it. If you’re someone who’s been curious about combining AI tools like ChatGPT and DALL-E with traditional design software like Photoshop and Adobe Character Animator, this is your sign to dive in.

Got any questions about the process? Feel free to reach out to me, and happy animating!