Inside D-ID’s ‘Digital Humans’ Build

Why are there so few startups taking the lead in Generative AI? Due to the high cost of training and operating Large Language Models, few Generative AI startups can operate without huge amounts of capital. A case in point is OpenAI — the maker of ChatGPT — which lost more than $540 million in 2022 alone.

However, thanks to the $13 billion in capital it received from Microsoft, I do not view OpenAI as a typical startup. For entrepreneurs seeking a piece of the Generative AI pie, such high capital requirements make it tough to get a slice.

As I work on my new book, Brain Rush, I have been eager to learn about startups able to hurdle those high capital barriers to entry. So I was pleased to speak recently with Gil Perry, CEO of Tel Aviv-based D-ID — the provider of a Generative AI-powered “digital human” — which has raised a relatively modest $48 million in capital.

Introducing the Digital Human

What is a digital human? “Generative AI with an AI avatar that can hear, think, and speak. No clicking, no typing. Just face-to face conversation,” according to a D-ID statement. Perry gave me a demonstration of a digital human that was able to take commands from Perry and talk, using facial expressions and hand gestures that appeared close to human. The demonstration made me realize this technology could take ChatGPT’s natural language prompts to a new level.

Why D-ID pivoted

D-ID has changed its business focus from the original idea. “When I thought of starting a company, I wanted it to be successful and do good,” Perry told me in an interview. “My first startup aimed to preserve privacy by protecting people against facial recognition. We entered Y Combinator and incorporated in 2017. We attracted top Israeli experts in deep learning and responsible AI — including Microsoft’s chief privacy officer.”

The rapid progress in LLMs motivated D-ID to build a new product to tap the growth opportunity. As Perry said, “When we saw how quickly people were developing LLMs, we decided to pivot from facial recognition to real-time, entertainment-quality digital humans in 2019 and 2020. We could use digital humans to create personalized customer support, brand ambassadors, and virtual sales assistants.”

What’s driving D-ID’s growth?

D-ID sees itself as making a big improvement in the way people interact with chatbots. “In 2024, all companies are using LLMs. We are replacing graphical user interfaces. We take human speech, convert it to text, get a response from the LLM, and convert the response to audio and video,” Perry explained.

D-ID sees a significant market opportunity. Gartner predicts the share of digital and marketing communications supported by avatars using text-to-video Generative AI will increase from 5 percent in 2022 to 70 percent by 2025.

D-ID — which has grown from 80 to 160 employees — does not seem to fear competition from technology giants. “If technology giants want to copy what we do, there is a big enough market for everyone,” Perry said. “It is a risk. However, it is hard to do what we do. We have the No. 1 platform in the space that gives developers the tools to customize digital humans through an application programming interface.”

D-ID is growing and is planning to move to Manhattan with help from its American team members. Perry attributes the company’s growth to its superior customer value. “We offer the best speed, the lowest cost, and the best quality. We have built a platform for developers that is scalable to the masses. We run fast and always need to stay ahead. We have top experts in deep learning, LLMs, and vision and they work together as a team.”

Will D-ID raise more capital and sprint to an IPO?

D-ID sees itself as being in no rush to raise capital. “We are very efficient and we still have most of the funds we raised,” Perry said. “Revenues are coming from product-led growth.”

After seeing a demonstration of D-ID’s digital human, it would not surprise me to see the concept proliferating in the years ahead. My question: How much of that opportunity can D-ID capture?