First revealed in February, OpenAI’s controversial yet powerful text-to-video generator Sora is now officially launched at Sora.com, not within ChatGPT. Initially, the tool was available to a select number of users, including visual artists, filmmakers, and designers; now, Sora is publicly accessible for no less than $20 a month as part of OpenAI premium subscriptions.
What is OpenAI’s Sora & how to create a video with it?
Sora is an AI-powered text-to-video generator that allows users to enter text prompts and turn them into videos. Incredibly realistic videos, as if created by humans, we must say. Apart from text prompts, users can convert videos out of photos and use Sora’s new Storyboard feature to string together multiple text prompts that Sora will then use to make several blended frames.
Sora — "A golden retriever, with a shiny wet coat, skillfully balances on a surfboard as it rides a gentle wave at Pacifica Beach. The dog's tongue hangs out in excitement, and its eyes are focused on the horizon. The backdrop includes the wide expanse of the ocean with rolling… pic.twitter.com/s5epk81RvB
— edwin (@edwinarbus) December 9, 2024
To create a video, users simply need to enter a text prompt and adjust the settings; four clips will be generated as a result. You can then make any video longer, blend it with another clip, change its ending, or replace objects within the video. With the Loop feature, you can create endlessly looping videos.
These are the generation settings users can customise:
- Aspect ratio: 16:9, 1:1, 9:16
- Resolution: 480p, 720p, 1080p
- Duration: 5 to 20 seconds
- Number of videos generated: Up to four at a time
- Video style templates: Options like "archival footage" or stop-motion
In addition to creating their own videos, users can also visit the Explore tab with Featured and Recent feeds constantly updated with creations from the community and add videos they like to Favorites; visit Library which is a collection of their own generated videos, and head over to Folders—a tool to organise videos into groups.
Sora also offers a Storyboard tool for generating videos in a consistent style and with related objects, which can later be combined into one larger video. Blend is another feature that allows users to combine two videos into one seamless clip, and Re-cut, another new tool, lets creators find and isolate the best frames, extending them in either direction to complete a scene.
OpenAI is launching Sora Today 👀
— Shane from Waking Digital (@digitalshane_) December 9, 2024
This was one of my favorite vids on the feed: pic.twitter.com/lVrhEKWpjr
Sora's current struggles & limitations
Sora is arguably the most powerful text-to-video generator as of now—the videos it makes are rightfully considered a real threat to human artists, that is how good the quality of the output videos is. However, the tool isn’t perfect at generating realistic physics and sometimes shows objects that vanish or pass through each other, according to the review of Marques Brownlee. Marques also noticed that Sora rejects prompts that feature public figures and copyrighted characters.
If Sora is so terrifyingly good, how are we supposed to know the video is AI?
OpenAI claims that the videos made by Sora will be visibly watermarked and have C2PA metadata that indicate the content is AI-made. Before you enter a prompt or upload a photo to turn into a video, you’ll need to check off an agreement to make sure you’re not uploading anything that contains people under 18, people who didn’t give their consent to be used in prompts, explicit or violent content, and copyrighted material. If you do, such “misuse of media uploads” can result in an account ban or suspension.
“We want to prevent illegal activity of Sora, but we also want to balance that with creative expression. We know that... will be an ongoing challenge, we might not get it perfect on day one. We’re starting to be a little conservative, and so if our moderation doesn’t quite get it right, just give us that feedback,” Sora product lead Rohan Sahai said during the livestream.
The Likeness feature is currently being tested with a small group of early users, since it allows the company to address concerns about misuse, particularly regarding deepfakes and the unauthorised use of someone’s likeness. To mitigate risks, OpenAI has implemented strict moderation standards for any uploads featuring people and continue to block content that includes nudity.
Users with access to the feature are shown in-product reminders about what is and isn’t allowed under the platform’s policies. OpenAI claims it actively monitors for misuse, removing inappropriate content and taking action when necessary.
“Today, we’re blocking particularly damaging forms of abuse, such as child sexual abuse materials and sexual deepfakes. Uploads of people will be limited at launch, but we intend to roll the feature out to more users as we refine our deepfake mitigations,” OpenAI press release says.
Besides, Sora uses algorithms that analyse users’ prompts and processes them if needed in order to avoid creating videos that resemble works of specific popular artists.
How to access Sora
The publicly released Sora is currently available only to OpenAI’s premium subscribers in the US and “most countries” except EU due to the regulations of the European Union. If you’re ready to shell out $200 a month for the OpenAI’s subscription, you can enjoy Sora’s video generating functionality and make videos in 1080p up to 20 seconds long.
The ChatGPT Plus subscription, according to OpenAI, allows you to generate 50 priority videos (1,000 credits) at resolutions up to 720p with 5-second durations. If you’re lucky enough to have a more expensive subscription, you’re able to download generated videos without watermark and generate up to 5 videos simultaneously.
