SYSTEM / DEC.2024
Finally, Sora is available! OpenAI’s video generation model, announced last year, is now available to the general public. After a 10-month wait, users can finally get their hands on this...
Finally, Sora is available. OpenAI’s video generation model, announced last year, is now reaching the general public. After 10 months of waiting, users can test it. Expectations were high. The result is frankly underwhelming.
The best part of Sora, for now, is its website. OpenAI has launched much more than a video generation portal. The company has built a community platform where users publish their creations and watch other people’s work. They can also create or edit videos with AI-assisted editing tools, prepare storyboards, or directly modify specific elements in a video.
This part works well. For a first public release, the integrated environment almost brings video generation into the realm of a more conventional editing tool.
The disappointment comes next. Once the novelty wears off, the technical results fall much further short.
Let’s be direct: Sora’s output remains below the expected standard. The demo videos published on the site can be impressive, with an obviously very carefully curated selection. Some of the showcased results probably required dozens, or even hundreds, of generations.
The price makes the experience even more frustrating.
$200 for 1 minute 30 of video. At that price, the tool becomes very difficult to justify for professional use.
The announcement of unlimited use for Pro users is also misleading. Once the credits run out, generation remains possible, but with major restrictions.
When creating a video, this pace brings everything to a halt. Rapid iteration matters enormously, because a usable sequence on the first generation is still rare.
Technically, Sora does better with abstract or fairly static videos. As soon as it has to animate complex objects, such as a running animal or a walking human, the results become frankly bad.
Sam Altman, OpenAI’s CEO, explained during the Sora keynote that the model is still in its early stages and is best seen as a GPT-1 version of video. The comparison doesn’t land well. GPT-1 was free and open source, whereas Sora can cost up to $200/month with limited credits.
Sora is an ambitious product and offers a glimpse of what video AI could eventually make possible. In its current state, its cost and usefulness remain difficult to justify. For the curious, the platform is mainly worth visiting to watch creations published by other users. For an investment of $200/month, I would definitely give it a pass for now. A more mature version needs to arrive before spending that money.
In summary, Sora still feels like a paid prototype.
EDIT - oh yes, something worth mentioning: of course, Sora is still unavailable in France for now. But it is available in Monaco, in case that helps. Go figure...