Pika Labs v1.5 AI video model is used regularly by over 11 million people. Many users use them to make fun and creative social clips. Smash faces, explode cakes, and more. In the last update, Pika introduced Pikaffects, which allows users to easily crush, melt, and explode items in an image. Used by brands such as Balenciaga, Fenty, and Vogue, as well as celebrities and influencers.

The new model still has that creative and social touch, but is more directly challenging the new powerhouse of AI video, OpenAI's Sora (#1 on the AI video leaderboard).

While OpenAI took 11 months to slowly build Sora, Pika's small team has created two new models and some impressive features. According to the company, the technical quality of Pika 2 is comparable to Sora, but offers better customization and control of output.

The biggest upgrade to Pika 2 seems to be how customizable it is compared to previous AI video models. In a statement, the company said this allows people to “create fun, entertaining, and engaging content.”

One of the new elements is the ability to control elements in the video by sharing your own images. This is called “scene material” by the Pika team and allows you to build a shot from the exact characters, objects, clothing, and settings that are in the image you shared with the model.

“The model's advanced image recognition intuits the role of each reference and combines them seamlessly into a single shot,” explains the spokesperson.

According to Pika Labs, the new model is also particularly good at following prompts. It follows both the intent and the details of the text and uses it to generate video. The company promises that “even the most complex prompts can be turned into clips without omitting or breaking down important elements.”

For any video model, the most complex element is movement. Even Sora, the best I've seen, struggles with particularly complex motion. I haven't tried Pika 2 myself yet, but from what I've seen in preview videos, they may have overcome that.

The company promises that its advanced understanding of physics makes real-life movements more realistic and fantasy-like movements, such as human flying or otherworldly elements, more believable.

I'm in the UK, so I don't have access to Sora yet, but I may be more excited about Pika 2.0 than I am about getting OpenAI's flagship video model.