Pika Labs v1.5 AI video model is used by more than 11 million people regularly. Many of its users turn to it for some fun, creative social clips. These include squishing faces and causing cakes to explode. Now, it's getting a huge upgrade with v2.
With its last update, Pika unveiled its Pikaffects, making it easy to squish, melt and explode items within an image. It has been used by brands like Balenciaga, Fenty and Vogue, as well as celebrities and influencers.
The new model still has those creative, social touches but more directly takes on the new powerhouse in AI video — OpenAI's Sora (which has gone in at number one in the AI video leaderboard).
While OpenAI spent 11 months slowly cooking Sora, Pika's small team has built two new models and some impressive features. The company says Pika 2's technical qualities rival Sora but offer better customization and control over its output.
What's new in Pika 2?
The biggest upgrade in Pika 2 seems to be how customizable it is compared to previous AI video models. The company said in a statement this would allow people to make "fun, hilarious, and engaging content they really want."
One of the new elements is being able to control elements within the video by sharing your own images. These are called "scene ingredients" by the Pika team and let you build a shot from the exact character, object, clothing and setting shown in the images you shared with the model.
"The model’s advanced image recognition intuits the role of each reference, and combines them seamlessly into one shot," a spokesperson explained.
According to Pika Labs, the new model is also particularly good at following a prompt. It follows both the intent and detail of the text and uses that to generate the video. The company promises that it "can turn even the most complex prompts into clips without omitting important elements or breaking down."
The most complex element for any video model is motion. Even Sora, which is one of the best I've seen, struggles with particularly complex motion. I haven't tried Pika 2 myself yet but from the preview videos it may well have cracked it.
The company promises it has an advanced understanding of physics, which allows it to make real motion more realistic and fantasy-like motion such as humans flying or elements on an alien world more believable.
I don't have access to Sora yet as I'm in the UK but I may be more excited for Pika 2.0 than I am to get my hands on OpenAI's flagship video model.
