How AI Saved Childhood Memories for the Trans Community
For a lot of trans people, pictures from childhood bring back uncomfortable memories. It’s not easy to look at a version of yourself you don’t recognise, or no longer are – especially when that’s attached to memories you cherish. To support the trans community by saving precious childhood memories, German NGOs Transklar e.V., Trans-Ident e.V. and Rosa Strippe e.V. partnered with Serviceplan Berlin in a transformative campaign.
‘Saved Memories’ features the use of artificial intelligence to transform the childhood photos of three people whose pictures no longer reflect who they are. Created by AI engineer Jaques Alomo through the AI tool Stable Diffusion, the original pictures were inputted and transformed to represent each person as their authentic selves they are today. As we see the reactions of social media influencer Duke Duong, cabaret artist Alice Dee and model and TV personality Lucy Hellenbrecht, the audience gets an insight into the impact of using AI in this way.
Serviceplan Berlin’s creative managing partner Myles Lord tells LBB’s Nisna Mahtani what it was like to create this powerful campaign, as well as the role that AI has in creating work within a limited budget.
LBB> When did the idea for this campaign first get mentioned? What was the starting point?
Myles> This idea came up almost a year ago as we were following the rise of AI in our industry and the new platforms and tools like Stable Diffusion that had come into existence.
LBB> Can you tell us a little bit about the non-profit organisations involved and the insight they brought to the campaign?
Myles> The three NGOs are focused on providing advocacy, support and counselling for the trans and Queer community in Germany. They were the ones who really believed in the power of the idea and were really interested in the initiative as a potential coping mechanism to help many trans people be more comfortable with their old pictures. The NGOs provided essential guidance and steering regarding the language of the campaign and how our protagonists are portrayed in the project. They played a crucial role - In this case, the NGOs were not just a client, but co-creators in the initiative.
LBB> Why were Lucy, Duke, and Alice the perfect people to take part and how did you find them? What did they think of the project and the main goals?
This was always going to be the most difficult part of the project – finding trans people that would be brave enough to show their old pictures and allow us such an intimate view into their lives. We reached out to many people that identify as trans on Instagram, mostly in Germany, but also all over the world. But we also spoke to a lot of ‘friends of friends’ to reach more people as trust would play such a big role.
Duke was someone that responded almost immediately. He was on fire for the idea and immediately confirmed participation. Then it was a while before we were able to sign up more participants. At one stage we even doubted if we could pull this off – but Duke was always like, “We have to do this people!” His confidence and positivity really keep us pushing forward. And really helped keep the project alive.
Alice came to us as a friend of a friend. And Lucy was a long shot. She is really famous in Germany as the first trans contestant in this big model casting show (the equivalent of Project Runway in the USA). We sent her management the presentation on a whim – and she said, “Count me in!”
These incredibly brave, beautiful, and inspiring participants were the other co-creators in the project. When we first pitched the idea to them, they really aligned with the aims of the project:
1. To pioneer a way to help trans people be more comfortable with old pictures
2. To create more trans visibility to foster more understanding and support for the trans community.
While we were in pre-production mode we talked about the emerging controversy in the US where Bud Light partnered with Trans testimonial; Dylan Mulvaney and received a horrifying transphobic backlash from celebrities like Kid Rock, and Travis Tritt in the press and on social media. Duke said to us, “If there are young people out there who are questioning who they are – I hope they don't see this hate and become afraid. I hope they see our video rather and know there are people around them understanding and supporting them.”
LBB> Let’s dive into the creative aspect of the campaign. What was the starting point of adapting the pictures using AI?
Myles> At first, we found the perfect partner in Jacques Alomo. A wealth of knowledge about AI and how it can be used to generate images and films. We met him at a training workshop and showed him the idea – he loved the aim of the project and began testing and experimenting.
LBB> Talk us through the process a little. What were some of the challenges you faced and how did you ensure the images authentically represented childhood pictures of the people who took part?
Myles> We all thought it would be easy – like you just click a button and hey presto it’s there! But this was way more complicated. AI is great at giving options. Hundreds and thousands of options. Even more, if you want – AI can just keep going. It’s like a police sketch artist interpreting someone’s description, except there are 1000 artists, and they will interpret your description in their own unique way. You still need the human eye to say: “That’s the one!”
The other thing we realised early on is that each photo was different and needed its own unique method or workflow to reach the desired result. Mostly what we did was train the AI with two data sets:
1. The original photograph where the AI should realign the gender.
2. Photos of the person today, which were simply made younger.
These two data sets were combined in the final generation, before it was fused with the original short once again. Here we used good old Photoshop as we normally would. In this way, we could make sure to hold on to as many details as possible in the original shot.
LBB> Why was Stable Diffusion the best tool for the process?
Myles> Jacques used Stable Diffusion because it has an API which allows it to work and interact with other software and apps. Jacques has also developed some software that sits on top to help with the training and prompting of his projects - to get the AI more on point. AI can generally be quite random – when you play with the tools it is often producing amazing results, but quite ‘off brief’ – it’s often hard to get it to produce exactly what’s in your mind. So in this way, Stable Diffusion is the perfect tool as we have a very specific task and here, we can really optimise our workbench for the job.
LBB> How long did it take to rework each picture?
Myles> Each picture posed its own unique challenge, so the times varied considerably. For example, Stable Diffusion had difficulties in some areas i.e., Asian faces. This simply means it was not trained so far with so much of that input, so we had to compensate. In some cases, we needed to first Photoshop to mask and isolate areas of the original shot to process individual elements i.e. Just the hair. So, the process was quite time consuming when everyone thought it would be just a few seconds in the magic box. The AI was amazing at giving us a broad range of options, but only a few options really capture the person and their identity. As I said before, the human eye was crucial in the process.
LBB> What was it like to see people’s reactions to their changed photos? And how did the trans community react to the campaign? Are you able to share some of the reactions with us?
Myles> We were quite nervous when it came to the reveal that you see in the film - as this was the first time the participants would see the pictures in many years. We wanted to really catch their true reaction on camera, so we kept the whole process away from them until they opened the book on the shooting day. They also had no idea which pictures we were using - these were provided to us by their lovely parents and grandparents. Original photo prints and slides hand delivered to us by courier. We had no idea how they would react. But the response was incredible. Many people are deeply touched when they see the proud look on Duke and Lucy’s faces. Alice was also really impressed, but not as much. But the fact that she could actually look at them was already a massive step. The three NGOs all agreed that this could be a meaningful coping mechanism to help trans people be more comfortable with their childhood pictures. By realigning these pictures with their true gender identity, we help return a missing part of their story.
We have had an incredible response from the queer community who have been posting this for pride month. We have been covered by the Daily Mail and big publications in Germany like Stern, Focus and Bunte. The reaction overall has been very positive. Some have said they are uncomfortable with the fact that we are changing the past. But what we discovered on this project is that we were not changing much. The adjustments are subtle - that’s why they look so authentic. And we are really adjusting the cues and gender norms placed upon them by society. Inside they felt this way all along. We just adjust the outside to reflect what was inside.
As we say at the end of the film, “Everyone has the right to be seen as they truly are.”
LBB> How do you think AI will impact campaigns like this in the future?
Myles> AI is going to be an incredible tool for non-profits, charities, and activists. In a space that relies on donations and volunteers you cannot afford many staff or any budget – and here AI will bring the chance to increase impact. AI is a way for us to automate labour-intensive tasks which means more power to us in achieving greater things. It will be a huge motor for innovation in social projects and social entrepreneurship.
