Twenty months is an eternity in AI. OpenAI, maker of ChatGPT, released its Sora AI video and audio generator in February 2024, and although the clarity of its image was often shockingly realistic, it wasn’t hard to notice it bending physics and breaking the illusion of reality.
Visuals would bend and teleport weirdly as Sora would “morph objects and deform reality to successfully execute upon a text prompt,” as OpenAI described. In launching Sora 2 on September 30, 2025, OpenAI said it has made a massive leap forward.
You can still tell something is off in a lot of the videos, but it certainly seems a lot more tightened up compared to the first iteration of Sora. The real concern, though, is that OpenAI is still going to suck up copyrighted material by default, so how many people will unknowingly end up feeding the beast that (potentially) eats their audience?
Play video
OpenAI sees Sora 2 as a large leap
“The original Sora model from February 2024 was in many ways the GPT‑1 moment for video,” wrote OpenAI in a September 30, 2025, blog post announcing Sora 2’s arrival. It added that compared to language-focused generative AIs, such as ChatGPT, video and audio generation models are in their infancy. Language-focused AIs still seem like babies, so then video-focused AIs must be downright fetal.
“With Sora 2, we are jumping straight to what we think may be the GPT‑3.5 moment for video,” OpenAI’s announcement continued.
Reuters reported that OpenAI representatives have said that copyright owners, including TV and movie studios, had to opt out of having their work appear in Sora 2’s video feed, which would be a policy carried over from its existing Sora image generator usage terms.
Opt-out permissions such as this, frankly, suck. They automatically scoop people up into complicity by defaulting them into “choosing” to allow their work to be siphoned up, merely through inaction or inattention.
Opt-in permissions, in which Sora 2 would default to copyrighted works not being included, would be much fairer.
You can already download the newly launched Sora iOS app for the Apple iPhone. Android users are out of luck, although you can access Sora 2 through Sora.com once you receive an invite.
Sora 2 is rolling out in the US and Canada, and it’ll initially be available for free, “with generous limits to start so people can freely explore its capabilities, though these are still subject to compute constraints,” as OpenAI puts it.
ChatGPT Pro subscribers can also use the “experimental, higher quality” Sora 2 Pro model on Sora.com. Although it’s not available in the Sora iOS app, Sora 2 Pro should be “soon.” If you fancy yourself a soon-to-be filmmaker, thanks to Sora 2, know that videos can only be up to 10 seconds long.