Sora from OpenAI is out: review, video examples, and limitations

Limitations

Sora is only available to ChatGPT Plus ($20 per month) and Pro ($200 per month) subscribers. The first ones are given 1000 credits per month, which is enough to create 50 videos up to 5 seconds long with a resolution of 720p. This is only enough for testing. To increase the limits, you need to switch to a more expensive plan. Then you will be able to generate up to 500 videos in the priority queue and an unlimited number in the general queue. You will also be able to create videos up to 20 seconds long with a resolution of 1080p. Videos can be downloaded, but with the Plus plan, there will be an OpenAI logo in the corner.

In addition, before starting work, you must agree to the service rules. Violations can lead to account blocking without a refund:

• You cannot use videos of people without their consent and videos of people under 18 years old.

• It is forbidden to upload media files containing violence and explicit scenes.

• All uploaded files must be copyrighted.

Also, with the $20 plan, you cannot create videos from uploaded media files containing people. The user is warned about this when logging in.

Sora is available in all regions where ChatGPT is present, except for EU countries. This is due to regulatory features in the EU.

Interface

At the bottom of the screen is the input field for requests and generation settings. You can select the aspect ratio (16:9, 1:1, 9:16), resolution (480p, 720p, 1080p), duration (from 5 to 20 seconds), the number of videos generated (up to four at a time). You can also choose a video style template, such as "archive clip" or stop-motion.

In the left corner, there are the following tabs:

• Explore — a feed of videos generated by users. Favorite videos can be added to favorites.

• Library — a collection of generated videos.

• Folders — folders to group your own videos.

Sora has Storyboard — a tool for detailed generation settings. With it, you can describe what should happen every second of the video. The mode is suitable for creating 20-second clips so as not to leave everything to the neural network's discretion.

Generating videos

For example, I asked Sora to generate a landscape of Moscow. I did not give any additional instructions on the specific place, angles, and conditions. Due to the influx of users, generating a 5-second video at 480p resolution took an hour and a half. This is considering that 480p resolution is marked as the fastest to create.

In the final video, it is immediately noticeable that Sora has its own idea of Moscow's layout. The postcard-like St. Basil's Cathedral ended up on the embankment, although it should be on Red Square itself. It is also visible that a road passes through the Kremlin square (if it is the Kremlin square), which does not exist in reality. The domes of the churches are scattered randomly and sometimes appear from ordinary buildings. The business center "Moscow City" is divided into two parts, located in the wrong part of the city, and its silhouette very remotely resembles the real "Moscow City".

If you have never seen the center of Moscow but have heard about the red walls, towers, and church domes, you might not even realize that something is wrong with the video. People familiar with Moscow will immediately notice the catch. It is also important to consider the rather low quality of the animations: cars and people look like random colored spots.

Video examples

Below are examples of videos from the feed generated by other users and the prompts for them. Each has noticeable flaws of varying degrees of absurdity.