Surreal AI Funding: How China's Synthetic Media Startup Grew

The Emerging World of AI-Generated Video

Imagine a future where video creation doesn't rely on cameras, but instead originates from a few lines of computer code. This concept, once relegated to science fiction, is rapidly becoming a tangible possibility.

Recent breakthroughs in machine learning are driving this transformation. We've already witnessed the capabilities of deepfakes, which can seamlessly swap faces or transform selfies into clips featuring well-known figures.

New Startups Pioneering AI Video Generation

Now, entrepreneurs with expertise in AI research are developing tools that empower users to generate remarkably realistic photos, voices, and videos through the use of sophisticated algorithms.

One such company at the forefront of this technology is Surreal, a China-based startup. Despite being only three months old, Surreal has successfully secured a seed funding round of $2-3 million from Sequoia China and ZhenFund.

According to founder and CEO Xu Zhuo, the company received close to ten investment proposals during this round, highlighting the strong investor confidence in the potential of AI-generated content.

From Snap to Synthetic Media

Prior to establishing Surreal, Xu dedicated six years to Snap, where he played a key role in developing its ad recommendation system, machine learning platform, and AI camera technologies.

This experience led Xu to believe that synthetic media would achieve widespread adoption, primarily because it has the potential to substantially “reduce the expenses associated with content production,” as he explained in an interview from Surreal’s Shenzhen office.

A Focus on Augmentation, Not Replacement

Surreal’s vision isn’t centered around replacing human creators or artists. Xu believes that machines are unlikely to surpass human creativity for several decades to come.

This philosophy is reflected in the company’s Chinese name, Shi Yun, which translates to The Poetry Cloud. The name is inspired by a novel from science fiction author Liu Cixin, illustrating the limitations of technology when compared to human artistic expression.

“Our core principle is that visual storytelling is a combination of creativity and production,” Xu stated. “We are concentrating our efforts on enhancing the production aspect.”

The Evolution of Video Creation Tools

AI-driven video generation can be viewed as an advanced video tool, representing a significant leap forward from the filters commonly found in today’s applications.

Short-video platforms like Douyin (TikTok’s Chinese counterpart) and Kuaishou have already lowered the barrier to entry for creating professional-looking videos, but they still necessitate the use of a camera.

“The true essence of short videos isn’t the format itself, but rather the superior camera technology that reduces the cost of video creation,” explained Xu, who co-founded Surreal with Wang Liang, a veteran from ByteDance, TikTok’s parent company.

The Commercial Application of Deepfake Technology

Leading technology companies globally, including Google, Facebook, Tencent, and ByteDance, are actively engaged in research concerning Generative Adversarial Networks (GANs). Xu’s approach diverges from directly competing with these industry giants, who typically pursue large-scale contracts. Instead, Surreal is focusing its efforts on serving small and medium-sized businesses.

Currently, Surreal’s software is exclusively available to enterprise-level clients. These clients can utilize the platform to modify faces within existing video or image content, or to generate completely novel visuals. Xu characterizes Surreal as a “Google Translate for videos,” highlighting its capability to not only swap facial features but also to translate spoken languages and synchronize lip movements with corresponding audio.

Service and Future Development

Billing is conducted on a per-video or per-image basis. Looking ahead, Surreal intends to expand its capabilities beyond facial animation to encompass the dynamic alteration of clothing and overall body movements. While specific financial details remain confidential, Xu reports that the company has processed approximately 10 million photo and video requests.

A significant portion of current demand originates from Chinese exporters engaged in e-commerce. These businesses leverage Surreal to produce marketing materials featuring Western-appearing models. The cost of hiring actual foreign models can be substantial, and utilizing Asian models has often proven less effective in these campaigns.

Customers utilizing Surreal’s “models” have reportedly experienced a 100% return on investment (ROI) in some instances, according to Xu. Bolstered by its recent multimillion-dollar seed funding, Surreal is actively exploring additional applications, such as within the online education sector, to gather extensive datasets for ongoing algorithm refinement.

• GAN Research: Major tech firms are investing in Generative Adversarial Networks.
• Target Market: Surreal focuses on small and medium-sized businesses.
• Key Feature: The software translates languages and synchronizes lip movements.
• ROI: Some customers have achieved a 100% return on investment.

The company’s strategy centers on providing a cost-effective and efficient alternative to traditional modeling and video production methods. This approach allows businesses to create compelling marketing content without incurring the high expenses associated with conventional practices.

Novel Ground in Synthetic Media

The core technology behind Surreal utilizes generative adversarial networks (GANs), a relatively recent advancement. First presented by machine learning expert Ian Goodfellow in 2014, GANs function through two key components: a generator responsible for creating images, and a discriminator tasked with identifying whether those images are authentic or fabricated.

These two components undergo a training phase characterized by competing objectives – an adversarial process – until the generator consistently produces convincingly realistic outputs.

The potential for misuse of GANs exists, with possibilities for fraudulent activities, the creation of inappropriate content, and other illicit applications. Consequently, Surreal is initially focusing on enterprise-level applications, rather than direct access for individual users.

Organizations such as Surreal are also introducing complex legal considerations. Determining ownership of machine-created images and videos presents a significant challenge.

To mitigate copyright infringement risks, Surreal mandates that clients possess the necessary rights to any content they submit for processing. Furthermore, the company embeds an encrypted, imperceptible watermark into all generated content, asserting its ownership and enabling tracking to deter abuse.

A potential concern involves the possibility of generated individuals resembling real people. To address this, Surreal employs an algorithm that compares newly created faces against a database of online photographs.

“Addressing ethical concerns is not solely Surreal’s responsibility, but we are committed to investigating these issues,” stated Xu. “At its foundation, synthetic media offers a transformative infrastructure, boosting efficiency. On a larger scale, its development is inevitable, as increased productivity is a primary driver in these scenarios.”