(Credit: Zain bin Awais/PCMag Composite;AppleStore)
Earlier this year, Google Gemini’s Veo 3 model brought realistic AI video generation with audio to the mainstream. However, Veo 3 now faces serious competition from OpenAI’s Sora 2, which offers the same capabilities. Here, I explain how you can access Sora 2, what it actually does, and if it’s any good.
What Is Sora?
Sora is ChatGPT’s AI video generation model, but it’s also accessible as a dedicated app separate from ChatGPT. It’s been around for a while, but it couldn’t generate audio nor match the quality of Gemini’s Veo 3 model until the recent Sora 2 update.
OpenAI didn’t just update Sora, though; it also updated the Sora app for iOS, which is basically TikTok for AI-generated videos: You can scroll through an infinite selection of 10-second-long videos that others have made with Sora, complete with comments and likes.
Where Is Sora Available, and How Do You Access It?
Sora is currently available on Android, iOS, and the web. Sora 2 was originally available only to those with an invitation, but wider access is rolling out at the time of writing. We suggest checking OpenAI's official social media channels for the most up-to-date information. According to OpenAI, Sora 2 is initially available for free with generous limits, but I suspect you will eventually have to pay for a premium plan to use it regularly.
If you pay for ChatGPT’s $200-per-month Pro plan, you get access to Sora 2 Pro, which is an experimental, higher-quality version of Sora 2. Pro users can generate 25-second-long videos compared with the normal 15-second ones. You create these longer videos through Sora's Storyboard feature, which lets you script out different scenes within a video.
If you enjoyed using the original Sora, Sora 1 Turbo is still available, along with all the content you generated with it. However, it’s unclear how long this will be the case. Considering the backlash OpenAI experienced after removing legacy models with the release of GPT-5, however, I expect Sora 1 Turbo access to stick around for a while.
What Can (and Can’t) You Do With Sora?
First and foremost, Sora generates AI videos. These can be pretty much anything, such as an anime scene you dream up or a fictional clip of a content creator playing Minecraft. Simply write a prompt, and Sora creates a video in about a minute or two. Once you generate a video, it exists as a draft that you can then publish, making it accessible in Sora’s TikTok-style feed.
Sora has a couple of other interesting features. For one, you can generate videos based on images you upload. You currently can't create videos based on images with real people, but OpenAI’s documentation strongly implies that capability is coming eventually.
(Credit: OpenAI/PCMag)In the meantime, you can generate videos of real people with the Cameo system. Essentially, you record a short video clip of your face and voice as your Cameo. Then, you can generate videos with yourself in them. Depending on your privacy settings, you can also allow other people to generate videos with your Cameo. For example, if you want to make a video with OpenAI CEO Sam Altman, simply include @sama in your prompt. For instance, “@sama explaining why Sora is better than Veo 3.”
Although Sora prohibits sexual content (unlike Grok's Imagine AI video model), there's still a small chance that Sora will generate it anyway. Similarly, Sora restricts any content that it deems violent. You can refer to OpenAI’s guidelines on what you are and aren't allowed to do with Sora for the specifics. However, these guidelines are somewhat confusing. For instance, even though Sora doesn't allow you to create videos of real people without their explicit permission, clips of dead celebrities and historical figures currently dominate Sora. Popular Sora videos include Bob Ross and Jesus, among others.
Is Sora Any Good?
I tested Veo 3 when it launched, and using Sora has been a very similar experience. In short, Sora 2's AI audio and video quality can be quite good. “Can” is the operative word, though, as it’s easy to encounter distortion and errors in your videos. That said, with a little prompt tweaking and multiple generations, it’s not difficult to make something impressive.
Compared with the original, Sora 2 is much better at rendering complex motion. For example, the original version couldn't generate a video of someone solving a Rubik’s Cube without major issues. Sora 2, though, does a significantly better job. The audio quality of Sora 2 videos is a bit hit or miss in my experience. Sometimes, it sounds great, but other times, the audio has a robotic, uncanny quality to it. It performs about the same as Veo 3 here.
If you use a Cameo to generate a video of yourself, the results don’t disappoint. It’s hard to describe the vaguely unsettling feeling of seeing a lifelike facsimile of yourself doing and saying things you’ve never done, but it’s undeniably impressive. However, Cameo-based videos aren’t free of the issues I mentioned above, so you still need to do some prompt engineering and run through multiple generations to make something really believable.
Sora’s App: AI TikTok
The Sora app is unique in the AI video space. (Veo 3 doesn't offer any sort of social component.) OpenAI is actively working to transform the Sora app into the TikTok of AI-generated videos. Some of these videos are undeniably amusing, and I appreciate the ability to comment on them as well. Still, it’s hard to appreciate that function too much when the majority of the comments are some variation of “wow” or someone begging for followers, at least at the time of writing.
(Credit: OpenAI/PCMag)I'm grateful that OpenAI created a social media platform for AI-generated videos, because it may help reduce the AI content I see on my TikTok and YouTube Short feeds. But, on the other hand, while some of the technology at play here is cool, I can’t help but wonder what the point of it all is: Is the ability to generate AI meme videos really worth building 60 football fields' worth of AI infrastructure every week or uprooting rural families?
Is Your Data Safe With Sora?
Since Sora is an OpenAI product, your data is as safe as it is with ChatGPT. In its privacy policy, OpenAI outlines how it collects significant amounts of data and uses that data to train its models by default. OpenAI doesn’t specifically mention Sora in its privacy policy, so there’s little reason to believe that it handles Sora-related data in a meaningfully different way.
How well does OpenAI secure the data it collects? Not very. For example, someone hacked OpenAI in 2023, news of which only broke in 2024, while OpenAI remains a target for hackers. Similarly, Sam Altman has said that your conversations with ChatGPT can be used against you in court, which isn’t very encouraging from a privacy perspective, even if it’s not surprising. In general, I don’t recommend sharing anything too sensitive with OpenAI.
When it comes to your facial and voice data within Sora, OpenAI tells PCMag that “uploaded Cameos are kept by OpenAI to use for new Sora videos, and are encrypted at rest and in transit.” It also says that “uploads are automatically deleted within 30 days after users delete them.” You can delete your Sora account, but currently, this is not possible unless you also delete your ChatGPT account. You also can't sign up for an account with the same email address after canceling an existing account associated with it. OpenAI is currently working on a way to let you delete just your Sora account.
Should You Use Sora?
Just as with Grok’s NSFW anime companions, you should check out Sora if you’re looking for a little entertainment and want to see what’s possible with the technology. Assuming you have an invitation, of course. Beyond that, though, it’s hard to see Sora as much more than a curiosity. Sure, you can create some meme videos to send your friends, but it likely only takes so many AI videos before you and your buddies move on to something more exciting.
Sora will also inevitably be weaponized: A world where you can create lifelike videos, with audio, of anything in just a minute or two for free is a world where seeing is not believing. So, I suggest never taking any video clips you see online too seriously, unless they come from a source you can absolutely trust.
Disclosure: Ziff Davis, PCMag's parent company, filed a lawsuit against OpenAI in April 2025, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.