AIFocus OnIn DepthStartups

Pika’s Valuation Jumps to 3.4B After Securing 80M in Funding for AI Video Tech

AI video industry leader Pika has returned with a new round of financing of 80 million US dollars (approximately 580 million yuan)! Investors include renowned institutions such as Spark Capital (leading the investment), Lightspeed, and Greycroft.

Post-investment, Pika’s latest valuation will reach 470 million US dollars (about 3.4 billion yuan). Founded by two Chinese doctoral graduates from Stanford University, the company launched its version 1.0 product at the end of last year and announced a financing round of 55 million US dollars. In the following six months, Pika expanded from an initial team of four to thirteen members, successively launched several new features, and was selected for the Forbes AI Top 50 list in April this year.

In the AI video generation track where Pika operates, both technology and investment have seen a dual explosion this year. On the technology front, OpenAI’s release of Sora set a new benchmark for the industry, extending the duration to 60 seconds. Subsequently, the open-source community released StreamingT2V, capable of generating 2-minute videos, followed by the domestic release of Vidu, which targets Sora’s capabilities with 16-second videos at a resolution of up to 1080P. Then Google launched Veo, which also targets Sora and can generate 1-minute videos.

On the investment front, following the surge in interest in text and image generators, the use of AI to generate videos is considered the next frontier, with both startups and large technology companies betting on this field. In addition to Pika, other AI video startups such as Runway and Synthesia have also raised a significant amount of funds in the past year.

Currently, startups like Pika are rapidly expanding because OpenAI and Google’s video tools have not yet been opened to the public, giving them a certain amount of time.

So, what has Pika been up to since the launch of version 1.0 half a year ago?

In the AI video generation track, what major events have occurred during this period?

What has Pika been doing in the past six months?

In summary, in addition to launching several new features that add to its offerings, Pika has been continuously updating its products and technologies. Here is a rough timeline of events:

In December 2023, shortly after the launch of Pika 1.0, its researchers proposed a method to increase the speed of text-to-3D model generation by 4.7 times.

This embedded acceleration algorithm, known as DreamPropeller, can optimize and enhance any score distillation-based text-to-3D model generation process.

Soon after, Pika announced its public availability, followed by the introduction of its paid subscription model.

By the end of February this year, Pika announced the launch of Lip Sync, a feature that can dubbing characters in the video and simultaneously “mouth” the lip synchronization.

Then in March, it launched the AI dubbing tool Sound Effects. With this new feature, users can generate sound effects from text or let Pika directly match the sound effects according to the video content.

By April of this year, Pika was selected for the Forbes AI Top 50 list less than a year after its establishment.

Forbes AI 50 List
Forbes AI 50 List

The above is the recent trend of Pika. Pika stated that the company currently has millions of users (the official did not provide an exact number in the Bloomberg interview), but it can be seen that there are nearly 800,000 users on Discord.

The AI video track has changed

A brief review shows that Pika has two founders, Demi Guo and Chenlin Meng, both of whom are Stanford doctors.

Looking at their academic experience, they are respectively focused on AI research in the fields of NLP and computer vision, and they both have academic experience in generative AI.

The two have a deep and solid professional background in the field of AI generation, so much so that they obtained 55 million US dollars in financing when the team only had four people.

Pika’s debut timing could not be more appropriate. Starting from November last year, within a month, five AI video generation products were released or received major updates.

Mainly including:

On November 3, Runway’s Gen-2 released a milestone update, supporting 4K ultra-realistic clear works.

On November 16, Meta released Emu Video, a tool that was said to have defeated Gen-2, Pika and other competitors in user evaluation.

On November 18, ByteDance released PixelDance, the dynamic nature of the work can be said to be unprecedented in history, and the elements in the picture are quite large in amplitude and do not deform, which is eye-catching.

On November 21, Stable AI also launched its own video tool Stable Video Diffusion.

Also on the 21st, Gen-2 launched the “paint where you move” motion pen refresh function, which is an important milestone in the controllability of the generation model.

Then came Pika’s official release of version 1.0 on November 29, challenging the “big brother” Runway.

It can be said that Pika was born when people were fanatically turning their attention to the field of video generation, and it caught the opportunity.

But no one expected that at the beginning of the new year, OpenAI’s “killer” video generation model Sora was released, which improved the quality of video generation to a higher level, and the duration was also extended to 60 seconds.

The heat of Sora has overshadowed everything, and later entrants to the industry also have to roll up their sleeves.

On April 7, GitHub open-sourced the video generation model StreamingT2V, which can generate a long video of 1200 frames and a duration of 120 seconds.

Following closely, on April 29, Tsinghua University and Shengshu Technology jointly released Vidu, the picture effect is in line with Sora, the length can reach 16 seconds, and it can also understand physical laws.

Then on May 14, Google released Veo at the developer conference, which can also generate a one-minute long video that targets Sora.

In short, AI video generation is becoming more and more mature in technology, and it has also driven the industry’s investment and financing trend.

In just one month in March 2024, three Chinese-founded AI video generation startups received financing, and the financing amount exceeded tens of millions of US dollars.

On March 7, Haiper, headquartered in London, UK (established by two former DeepMind team Chinese engineers), announced the launch of the product and had completed a financing of 13.8 million US dollars.

On March 13, Aishi Technology, founded by the former head of ByteDance’s visual technology, completed a round of A1 financing of about 138 million US dollars (about 13.8 million US dollars).

On March 24, HeyGen, an AI digital human product that made Taylor Swift speak Chinese and became popular (founded by former Snap software engineer Xu Zhuo), completed a round of financing of 60 million US dollars.

On April 23, Aishi Technology completed a round of A2 financing of more than 100 million yuan, led by Ant Group.

On June 1, Runway obtained a D round of financing of 100 million US dollars, led by Google.

On June 5, Shengshu Technology announced the completion of a pre-A round of financing of several hundred million yuan, led by Beijing Artificial Intelligence Industry Investment Fund and Baidu.

It is not difficult to see that AI video generation will usher in a new outbreak this year, and as one of the top streams, Pika naturally attracts attention.

And Pika now claims that they have built their own “foundational model”, so they will not rely on the technology of other AI companies.

Although a lot of funds are needed to train the model, the founder Demi Guo revealed that Pika does not need to consume a lot of funds in this area. They are a bit similar to Mistral AI in France, and they can also train a relatively good model with less funds, but the specific details have not been disclosed.

Finally, today is Pika’s first birthday (calculated from the day the founders took a break from school to start Pika), and the official also released a commemorative video.

Next, Pika plans to release a major update to its text-to-video model later this year, which includes two aspects: one is to provide higher quality and longer videos, and the other is to ensure the consistency of the output.

According to the founder Demi Guo, Pika’s new tool can allow users to define an object or role, and then generate it consistently in the video or finally in different clips.

Ella Shi

Ella Shi is a third-grade student of the Literature Department, a lover of literature and photography. She used to work for the college media community as an author of the school paper. Now she joins FirmKnow to practice and explore her business knowledge.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button