Deal focus: Tiamat innovates around words and pictures

Prior to the emergence of ChatGPT and its artificial intelligence (AI) chatbot capable of engaging in human-like conversation, the focal point of much fascination in the AI-generated content space was “text-to-image” generation.

In January 2021, OpenAI – the Microsoft-backed research lab responsible for the large language model (LLM) technology that underpins ChatGPT – launched Dall-E, which uses natural language descriptions, or prompts, to generate digital images. An update, dubbed Dall-E 2, was released in April 2022, offering more realistic images at higher resolutions.

Competing labs were quick to respond. Midjourney launched its debut text-to-image product in March 2022 and followed up with new iterations capable of higher-quality generation. Stable Diffusion arrived in August of the same year and its platform has since accumulated more than 2,700 projects, almost ten times the Dall-E 2 total.

From the beginning of 2022, DCM China tracked a series of interesting AI-generated images posted on local social e-commerce platform Little Red Book. They were the work of a college student who had developed a solution derived from Stable Diffusion and Dall-E. DCM committed USD 1m and the founder quickly built up a team that started working on its own model based on the underlying diffusion infrastructure.

“If you want to create an entry barrier and provide unique value add, you need to write your own version,” said Hurst Lin, founding partner of DCM China, “This version came into being after we provided funding.”

The start-up – known as Tiamat – closed its Series A round close to USD 10m, co-led by DCM and Vitalbridge Capital, last November although it wasn’t made public until last week. Jinjian Zhang, Vitalbridge’s founder, said he was won over by Tiamat’s ability to leverage AI to improve work efficiency.

“If you combine the diffusion model and Controlnet, a tool that controls and adds extra conditions to image generation, you can achieve 40%-50% of the effects delivered by Adobe Photoshop. On this basis, we thought it was a good time to look at AI-generated content (AIGC) in China at both LLM level and application level,” said Zhang. “We have 5 or 6 other companies in the pipeline, in addition to Tiamat.”

In the final quarter of 2022, Tiamat claims to have received orders amounting to several million renminbi from anime, advertising, and design companies. “These customers have a very simple demand: they want access to more advanced tools to help them solve productivity problems,” Zhang explained.

Tiamat’s model taps into multiple functions: text can be used to guide the generated image or a sketch is utilised to deliver a detailed result. The company’s text-to-image model has billions of parameters – a huge number but actually 100x smaller than ChatGPT, which means the training cost is much lower.

“Regarding the difficulties of data processing, the easiest aspect might be image generation. It becomes more complicated with dialogue, especially multi-round dialogue because it involves the concept of time, so there is an extra dimension. The next step up is 3D and video-like content generation,” said Zhang.

Tiamat’s competitive advantage is in specific scenarios rather than model size. In apparel, for example, AI can replace or help designers provided the machine understands industry terminology around patterns and design. Tiamat studies communication between designers, and the corresponding images, and then adjusts its model accordingly.

ChatGPT-related start-ups are proliferating in China, but Lin believes AIGC has become overhyped. He warns that the market could be distorted by opportunist investors and entrepreneurs who don’t really think through use-case scenarios.

“In the short term, there are too many copycats, too many wannabes, and too many people who just one year ago were talking about how web3 was going to change the world and are now switching completely to generated AI,” Lin said. “In the long term, once the hype dies down, I think we will have our next ‘internet revolution’ with very important use cases emerging and technology catching up.”

• Tweet  
• Facebook  
• LinkedIn  
• Google plus  
• Save this article  
• Send to  

• Topics
• Greater China
• Early-stage
• Technology
• China
• Vitalbridge
• DCM-Doll Capital Management
• artificial intelligence