Superhero Transformation: An AI Adventure in Employee Onboarding

“Great, won’t take 30 minutes,” I naively thought to myself. Oh, how wrong I was! What started as a simple idea to spice up our employee onboarding process turned into an exciting journey through the landscape of AI capabilities, teaching us valuable lessons along the way.

The Mission

At Enablis, we believe that first impressions matter. When reimagining our onboarding experience, we wanted to create something special – something that would make new team members feel like the superheroes they are from day one. One idea in this space was to generate personalised superhero avatars for each new employee, for things like Slack. Afterall… What if your first day on the job turned you into a superhero? 

Simple, right? Well, buckle up for this roller coaster ride through the world of AI!

Act 1: The “Simple” Solution

Armed with the power of modern AI image generation, our first thought was to leverage tools like DALL-E through ChatGPT. “We’ll just create a Slack integration,” we said. “It’ll be seamless!”

It was not seamless.

Slack’s evolution had introduced new barriers to integration. What was once a simple webhook had become a complex dance of permissions and app configurations. Back to the drawing board!

Act 2: Keep It Simple… or Try To

Plan B? A no-fuss website to do the heavy lifting. A static website hosted on S3 with some JavaScript magic. Users could input their hobbies and interests, hit submit, and voilà! Except… not quite.

Our first backend implementation was classic AWS: Lambda behind API Gateway, calling OpenAI’s DALL-E. The results? Well, let’s just say our superheroes looked more “generic background character” than “personalized protagonist.” but why?…

Act 3: The Plot Thickens

This is where things got interesting. We needed our superheroes to actually resemble our team members, but we ran straight into one of AI’s most important ethical guardrails: the intentional limitation on generating likenesses of real people.

You see, OpenAI and similar providers have implemented strict protections against generating images that could be used to impersonate specific individuals. Even when we tried feeding in photographs with explicit permission from our team members, the AI politely but firmly refused to generate images that closely resembled them. It’s a crucial safeguard against potential misuse – think deepfakes and identity theft – but it meant our superhero dream needed a creative workaround.

Our attempts included:

  • Image alteration APIs, taking existing profile shots and re-imaging them (Result: Shocking, and not in a good way)
  • Base64 encoded image prompts, asking the AI to describe the image payload (Result: Nonsense, total guesswork! A banana somehow generated a random persona – yes, really!)
  • Direct image attachments to ChatGPT (Result: Met with clear ethical boundaries, scene details not personal details were instead provided).

Not even a yellow suit…

Each attempt taught us something about the delicate balance between innovation and responsible AI use. While our intentions were purely fun and consensual, we were asking these models to do something they’re specifically designed to avoid. It’s also fair to say that some of the early generations were less flattering than others!

Looking very sinister!


The Final Form

After much experimentation, we landed on an ethically sound compromise: instead of trying to directly replicate appearances, we built a system that captures general characteristics while maintaining enough creative distance to avoid any ethical concerns:

    1. A user-friendly S3-hosted site behind CloudFront.
    2. API Gateway triggering Step Functions for orchestration.
    3. AWS Rekognition identifying broad characteristics like gender, age, and facial hair (with explicit user consent).
    4. Claude API analyzing general features like hairstyles, but in an abstract way.
    5. All these characteristics feeding into our “Creative Brief” for final image generation, with enough artistic interpretation to create something unique rather than replicative.


      The result? Superhero avatars that capture the spirit of our team members while respecting both privacy boundaries and ethical AI guidelines. They’re recognizable enough to be personal, but stylized enough to avoid any concerns about deepfakes or identity replication.


“You wish” was my wife’s reaction… Here’s the app having generated an Avatar.


Lessons Learned

This “30-minute project” taught us valuable lessons about the current state of AI:

  • Ethical boundaries in AI aren’t bugs – they’re features protecting us all
  • The gap between what’s technically possible and what’s ethically responsible is a crucial space for innovation
  • Sometimes limitations force us to find more creative solutions that end up being better than our original idea
  • The industry can rapidly deploy AI models, but “One size fits all” approaches often fall short for specific use cases
  • Combining multiple AI services can create sophisticated solutions that respect both user needs and ethical guidelines

Future Power-Ups

Looking ahead, there are several exciting possibilities to enhance our superhero generator:

  1. Style Consistency: Improvements to our prompt and potentially training a fine-tuned model specifically for superhero avatar generation could provide more consistent results while maintaining individual likeness
  2. Likeness: Having an image generation API that could consume image data and use that in the creation of avatars would bring much better options for achieving greater likeness.
  3. Interactive Customization: Adding real-time preview capabilities for costume and power selections. A feedback loop into the prior conversation with the AI would allow this to be interactive.
  4. Animation Integration: Exploring ways to bring these avatars to life with simple animations. 
  5. AR Integration:The next stage could see us create meshes and render avatars in something like Unreal for use in the AR space. 

The Secret Identity

While our journey took significantly longer than 30 minutes, it revealed something important: sometimes the most valuable outcomes come from embracing the complexity rather than fighting it. What started as a simple idea evolved into a sophisticated system that balances technology, creativity, and ethics.

And isn’t that what being a superhero is all about?

Want to join our league of tech superheroes? Check out our careers page – cape optional, powers guaranteed!