Familiar Faces: Creating "Normal" People using Gen AI

James Dow

Rethinking Advertising: Normal People

For years now, the advertising world has steered away from conventional and idealised standards of beauty. The intention, instead, has been about crafting campaigns that resonate with the diverse realities of everyday life. Authenticity is the new north star, with campaigns increasingly seeking to connect with audiences on a genuine and relatable level. It’s why UGC is so important now, and it’s why brands are looking to smaller influencers to help them drive revenue. People trust people who look like them.

The Creative Challenge

The advertising world has long grappled with the notion that generative AI struggles to capture the essence of "real" people. If that is indeed the case, then AI will struggle to meet expectations in this new landscape where authenticity is key. If AI is to compete with in-person shoots and street casting, we need to see it produce images of people that could be your friends, neighbours, colleagues. Those people need to feel real.

Street casting has become a powerful trend in advertising, with campaigns platforming real people rather than professional models. It’s a way of accurately representing the myriad of faces within the communities that brands want to address. The challenge for generative AI, therefore, is to replicate the spontaneity, individuality, and authenticity that you’d find in those real-life encounters. 

Creating those relatable characters with Gen AI is a pivotal step if AI can keep up with marketers’ needs. We need to see AI helping to create more inclusive and genuine campaigns, with characters that resonate with a broader audience.

The Outcome

We wanted to see if Gen AI could create images of people that felt real. We wanted to see if you could get a sense of the person’s character, without any hints from their clothing or the setting they’re placed in. We looked at creating four different characters in plain clothes in front of a plain background. We were able to create four people of various ages, races, body types and facial expressions. 

The AI was able to create more serious expressions as well as all manner of smiles - cheeky, proud, soft. Above all, though, they’re relatable people. You can easily imagine the adults as police officers, teachers, software engineers. As for the young girl - if you saw her image in a school prospectus, you wouldn’t think twice about it.

The Methodology

Made by @JamesD0w (Creative Director, Brandtech Consulting). Here’s how he did it:

1. Prompts

  • A portrait photo of a 40-year-old, black woman standing in a grey background photo studio. Sweatshirt. Plain face, neutral emotion.
  • A portrait photo of a 40-year-old, white woman with wavy blonde hair, standing in a grey background photo studio. tired. Smiling, wrinkles, high definition skin texture.

2. Images

Super quick one here. It’s really all about the prompt. Gen AI generally is fairly faithful to what you ask for. However, if you ask for a ‘woman’ you will most likely get a biased image generated. There are many reasons for this but the main point is that we can mitigate for this by being very specific.

Take your time to engineer the prompt thoroughly, work out how the model is reading your words, and try to find the right keywords for your character. It’s about testing and improving.

3. Upscaling

Finally, we upscale all the preferred images using Magnific AI on the following settings:

  • Creativity 1
  • HD 5
  • Resembalnce -1

The Conclusion

We found that Pencil’s AI was more than up to the task of creating images of authentic, relatable people. Without relying on clothing, props or settings, we were able to create people that you could easily envision within local communities. If brands want to keep up with high-performing UGC-style content, they need to include relatable people in their campaigns. The difference now, though, is that you don’t need to take to the streets to achieve that authenticity.