AI video-generation start-up Pika raised $80 million as venture capital investors continue to bet on the small teams in the fight with industry heavyweights.
Stanford PhD students Demi Guo and Chenlin Meng founded Pika a year ago, serving as CEO and CTO respectively. Initially, they launched a simple product on Discord, which quickly gained popularity and led to a $55 million seed investment. Investors included former OpenAI founding member Andrej Karpathy, Perplexity founder Aravind Srinivas, and HuggingFace founder Clem Delangue. This investment was more of an endorsement, a stamp of approval for this young team.
Pika then released a publicly available product. Afterward, Sora emerged. All AI video companies faced a common challenge: how to avoid being outdone by Sora. Today's funding round, led by Spark Capital with participation from Greycroft, Lightspeed Venture Partners, and others, indicates that Pika has found the answer.
Here are the key points from our recent conversation with Pika.
Pika is not Sora
Pika has its own foundational model, which gives them a competitive edge by not relying on other base models. Pika's founding scientist Omer Bar Tal believes that with a powerful foundational model, the basic capabilities and functionalities of video generation are strong enough. Continuous iteration of the model can bring more possibilities to the product.
The Pika team was initially anxious about Sora's emergence. However, after some observation, they concluded that there is currently no concrete evidence of Sora's true capabilities, mostly relying on engineering efforts and luck. In terms of product plans, they found that Sora does not seem to have plans to launch a product truly aimed at the public in the short term.
The Pika team believes there is a significant difference between "showing the five best demo effects" and actually providing users with large-scale usable products. Sora's demos require enormous human and computational resources, which results in different video content when these resources are not available.
For Pika, the priority and goal have always been to provide users with better features, not just showcasing demos.
Therefore, for Pika, the more important research direction is how to integrate powerful models with the product.
They believe the model is already an important part of the product, but many features need real users to discover. While maintaining talent density, Pika frequently collaborates across teams internally to ensure straightforward communication with a more tightly-knit and flexible setup. Typically, engineers and researchers upgrade the model's capabilities, then discuss with the product team to identify features from a user perspective and then materialize these features.
In February 2024, Pika launched the Lip Sync feature, which many competitors lack. For ordinary users, lip-syncing has always been a must-have feature and videos made with this feature could easily go viral. This product design requires extensive data to fine-tune the model for the corresponding capability. Pika subsequently launched Sound Effects, adding sound effects to videos. These features are continuously updated, offered through both free and paid modes to meet different user needs.
Pika is also gaining ground in business customers, indicating potential growth in revenue.
A TikTok for creators
Unlike those hardcore tech companies, Pika's team includes a creative department that explores Pika's product capabilities, representing Pika's primary target users—individual creators.
"Everyone can be their own director." This is Pika's slogan.
This positioning is noticeably different from Sora, Runway, and other products. These more heavily funded products often showcase works and application scenarios leaning towards Hollywood-level collaborations. There are reports about OpenAI signing deals with Hollywood, attempting to enter this massive industrial system. Pika, however, aims to capture the individuals.
The Pika team believes that in terms of creative significance, the creative liberation TikTok brought to individuals is also evident in Pika. Perhaps the difference with TikTok is that Pika offers higher creative quality.
So Pika is more like a TikTok for creators.
In fact, TikTok has changed how people interact with video, and with generative AI, this interaction can be redefined again. Demi Guo often mentions a long-term goal within Pika: to completely reshape the inference of video production and consumption through new technologies.
Moreover, the concept of TikTok for creators offers more imaginative possibilities for Pika and provides new ideas for addressing today's business model challenges in the generative AI field. Beyond project-based deployment, deep customization cooperation, and subscription fees, the platform economy can open up more possibilities.
Next, Pika's new user interface will go live, making it simpler and more user-friendly. Meanwhile, Pika 2.0 will see major updates at the model level. From the product to the video model, the next phase will bring an entirely new Pika.