Hey there,

OpenAI has announced its next generation video model, Sora 2. This new model builds on the original Sora model that was released late last year and brings a higher level of creative control and video quality. OpenAI described this as a jump from GPT-1 to GPT-3.5 in a single year. This week we’ll look at what specifically has been improved and how it compares to other video generation models currently on the market.  

What’s New?

While Sora 2 brings improvements across the board, I want to focus on a few of the major upgrades. 

Higher Quality:

One of the first things you’ll notice when looking at a Sora 2 video is the incredible quality. This was the biggest “wow” factor for me because I was able to generate incredibly detailed videos on my first attempt. Things like more realistic lighting, particle effects, and physics help make your videos feel real.  The audio quality has also been improved in Sora 2 as videos now include sound effects, music, and even dialogue. 

Improved Character and Style Consistency:

One of the most important (and irritating) issues with video generation is inconsistencies between videos. While making a single 5 second video is cool, being able to create a coherent story between several of them is truly impressive and shifts AI video generation from a novel idea into a practical tool. OpenAI created Cameos to quickly and easily allow us to generate videos with preset or saved characters. These characters can be entirely original creations, based on you, or even based on other people. 

Stitching and Combining Videos:

As we’ve seen with other video generation platforms and models, Sora allows you to extend your original creation by combining it with other clips. You can do this by selecting multiple videos and clicking “Stitch”. While this is still far from a video editing tool, it serves as a simple way to combine creations, enabling you to build out stories and see your ideas come to life.

The Platform:

It's impossible to talk about the Sora 2 model without mentioning its new home: the Sora app. OpenAI has created this platform as an AI-only ecosystem for its videos. While you can keep all your creations private, the goal is clearly to build a community and share with your friends. This integrated approach of model and platform together expands OpenAI beyond ChatGPT, positioning them as a potential social media company. We won’t be diving any deeper into the social media side of the platform but it has potential to be an interesting place to explore. 

Safety & Access

One of the biggest challenges around AI video generation is that it can be hard to identify it. OpenAI has taken steps to help alleviate this: adding a watermark to all videos and hosting these videos on the Sora platform. The watermarks allow people to quickly identify Sora 2 videos wherever they may be posted. The platform itself being only AI content means people won’t take what they see at face value and instead just enjoy the wacky creations. 

Access is currently restricted to the US and Canada but will be rolling out globally soon. To get started, you need to download the Sora app and create an account. Once you have an account, you can use the web version for further creation. 

My Thoughts: 

OpenAI has taken an important step with Sora 2. Not only have they expanded with the Sora social media platform, but they’ve also created a competitive state-of-the-art video generation model. While the social platform is a significant part of this launch, my focus here is on the underlying model and its creative potential. The model itself is really exciting and I’m not the only one who thinks so. Sora 2 is currently tied for first place in the LMArena leaderboards with Google’s Veo 3 models. I’m someone who’s extensively used Veo 3 and I’ve become reliant on the features included within Google’s Flow so it's nice to see that OpenAI has created a Sora-powered competitor. While the lack of editing tools is a bit disappointing, Sora 2 gets me excited to see what OpenAI creates next!