The campaign of the International GenAI Conclave in Kochi, Kerala created several fictional AI characters including Ramankutty who is a farmer.
In the days leading up to July 11th’s launch of the first ever International GenAI Conclave held in Kochi, Kerala and organized by the government of Kerala State in collaboration with IBM, an innovative campaign broke into public domain both online and offline.
There was a well-arranged campaign that had four characters originating from different backgrounds within their society. They were Josettan representing ordinary people seeking help from local authorities; Ramankutty: farmer; Usha Chechi: tailor and Minimol as a budding businesswoman.
Even experienced cameras were deceived by these figures and their backdrops which looked so human like that they assumed it was someone’s picture instead. Yet they were all generated through AI.
“We wanted to make AI more accessible so we associated it with local lives and livelihoods rather than being highly technical, complicated and intimidating as it is generally believed. This led us to create the slogan for our campaign ‘AI for the People,” says Shelton Pinheiro, Creative Director at Stark Communications based in Thiruvananthapuram that executed this advertisement.
Initially, making Keralite-faced figures was daunting due to software used for image creation that mainly produced north-Indian faces whenever India-related prompts were given. Additionally, India-associated visuals were mostly “poverty-laden.”
“Now, with Indian users on the rise, the algorithm returns more specific features such as that of South Indians on proper prompt. But even then, with the ever-increasing volume of data, even the very same prompt may not necessarily return the same results,” said Mr Pinheiro.
To get accurate prompts one can also use reference images when typing them in many ways including uploading them. Moreover, when there are details about camera set up or scene specifics contained in prompts; the backdrop can be accurately created.
However, with third-party software that facilitate upscaling by increasing the number of pixels, bigger outdoor images can now be made using AI unlike before. It is even harder to differentiate AI-generated images from real ones due to more advanced software which poses its own challenges.
So how do we know these four AI-generated figures do not look like someone in reality? While conceding that there may be a slight possibility, he argued that AI generated pictures by “scouring and scrubbing” millions of images from the internet hence making this unlikely.
The Industries department along with IBM loved the campaign idea. “The aim was to create intrigue around the campaign, and it seems that has been achieved. We also got very good feedback on our campaign materials kept at the venue and along the way,” said Mr Pinheiro.