When OpenAI published a demo of Sora back in February, the hype train made its way across all the major tech blogs, carrying news of Sora’s potential. After months of anticipation, on the 9th of December, OpenAI finally released it to ChatGPT users on paid plans (but only in select regions).
I happened to catch wind of the release that same day. Unfortunately, I didn’t catch it early enough: by the time I went to try it, OpenAI’s servers were so overloaded with Sora frenzy, that they had to temporarily block access to new users. I then forgot about it for about two weeks but it came back across my radar while I was putting together the latest issue of our Bizarro Devs newsletter. I took it as a sign that it was time to finally jump in and take Sora for a test spin.
If you haven’t tried Sora yet (or are in a region where it’s not currently available) and you’re curious about my results, then keep reading. I will begin with an overview of Sora’s usage limits, its UI / UX, and the different ways you can make a video with it. Then I’ll share samples of the videos I made – both good and bad – and I’ll review some of Sora’s additional features that you can use to edit your videos.
Sora usage limits ⚠️
When you first log into Sora, you are greeted with a popup window that shows you the user limits of the particular OpenAI subscription you’re on. In my case, that would be ChatGPT Plus.
As you can see below, on the Plus plan you get 1,000 video credits per month, while the Pro plan gives you ten times that amount. Seeing as the Pro plan costs exactly ten times more than the Plus plan, the math adds up here:
You’ll also notice that it says the 1,000 credits can theoretically let you create up to 50 videos. In practice, it highly depends on what other variables you tweak when creating your videos.
Realistically, based on my experience, 👉 most people will burn through those 1,000 credits way before coming anywhere near close to 50 videos.
For context, I used up my first 600 credits by making 13 videos at the lowest possible resolution (480p), with most of them being ten seconds long. Assuming I continued at that pace without any adjustments, I would have run out of credits before getting even halfway to 50.
Fortunately, you can still continue making videos even after you run out of credits. The main caveat is that your requests will be added to what OpenAI calls the “relaxed queue.” This just means your videos will take a longer time to generate. You’ll also be capped at a maximum of two simultaneous video requests (which is lower than the normal ceiling when using your credits).
The Sora UI 💻
I have to give credit to the designers at OpenAI who worked on the user interface because they made it extremely accessible. Compared to video editing programs like Davinci Resolve or Final Cut Pro, it’s on the complete opposite end of the learning curve spectrum. The basic functions are that straightforward.
The main page looks like what you see below, with the prompt window being found at the very bottom of the screen:
In addition to the prompt text area, there are nine buttons there. Let’s go over them:
Understanding the prompt window options
Starting from the far left, this is what each of the buttons let you do:
- Upload an image or a video (or choose from previously uploaded images and videos).
- Select from five video style presets to give your output a certain look and feel. For example, you can go for Film Noir, Archival, and more.
- Pick the framing orientation of your video. There are three choices – wide, square, or tall.
- Choose the resolution quality. This one also has three choices – sort of. While the popup menu does show 480p, 720p, and 1080p, I was only able to select 480p or 720p. The 1080p is reserved for ChatGPT Pro users.
- Decide the duration of your video – 5, 10, 15, or 20 seconds. In practice, I was only able to choose either 5 seconds or 10 seconds. Additionally, when choosing 10 seconds, it wouldn’t let me increase the resolution to 720p. Put another way, 480p videos can be either 5 or 10 seconds long, but 720p videos can only be 5 seconds long.
- Based on your prompt, generate either one video, two videos, or four videos. For 720p resolution, you can only do a maximum of two videos at a time. For 480p you can do four.
- Hover over the question mark and discover how many credits you will use to generate your video based on the input parameters from the previous buttons.
- Open up the storyboard page. This is where you can go into depth by breaking up your prompts into individual scenes. You can also upload photos or videos here and add them to a scene.
- Create your video based on the prompt you put into the “Describe your video…” text box. If you just want to generate a video as fast as possible, then you can skip the storyboard and send your prompt straight to the queue for generation.
Understanding the pathways to make a video 📽️
As you just read above, there are multiple ways you can generate a video. You can also mix and match certain elements. For example, you can:
- Write a prompt and tap the far right ⬆️ button to create your video based on the prompt. This is the fastest option.
- Upload a photo, add a prompt, and tap the ⬆️ button.
- Upload a photo and tap the Storyboard button to go to the Storyboard page. Sora will automatically analyze your uploaded photo and provide a prompt for a scene based on the content of the image. You can also edit the text of the auto-generated description. In the example below, I edited the text by adding that the specific location of the image is Chinatown in Bangkok. I also added the time period.
- Write a prompt and tap the Storyboard button. Sora will take your prompt, extract the text, and generate some scenes from it.
Assessing Sora’s output 🎬
Now that you have a general overview of Sora, you’re probably wondering if the videos it makes are actually any good.
To put it bluntly:
They’re mostly unusable.
Out of the 21 videos I generated in total, only 7 of them didn’t contain any massively visible errors. The majority of them were akin to nonsensical digital slop. Below are three examples to give you an idea. Two of them I showed you the prompts for in the previous section.
This doesn’t need a whole lot of commentary from me. It’s obvious how bad these are. And these were the norm, not the exception.
Even bigger problems
On top of giving me mostly bad output, I had two other issues:
- One video that rendered as nothing. I uploaded an image and opened the storyboard. Sora analyzed the image and provided a description of the scene. It looked good to me so I tapped the ⬆️ button, but when I received the output, it came back as the same still image I had uploaded.
- On a different attempt, I tried to render a video using an uploaded photo as a starter. OpenAI’s system blocked me.
There was nothing controversial about the photo. The only thing I could think of that might have possibly caused it is that there were humans in it.
Sora explicitly warns you that you cannot upload photos or videos of humans.
Although I was aware of that when I added the photo, in my opinion it was an edge case. Yes, there were humans in the photo, but none of their faces were clear. It was an old photograph from the mid-20th century and the focus was more on the overall location than any individual person.
On the plus side, before showing me the error message above, it did give me the auto-generated scene description of the photo. I was then able to take that and paste it back into a fresh prompt. The result was one of the few decent videos I managed to get out of Sora. Speaking of those decent videos…
The decent videos
Since I shared three bad video examples with you, it’s only fair to share three good ones as well.
Turning bad videos into decent ones 🛠️
I have a confession to make: there’s something I haven’t told you yet.
While the initial output of most of the videos was unusable, Sora has two tools that let you clean up some of the undesired elements within your videos. Each of the tools functions via different mechanisms. Let’s go over them.
The remix feature
The most direct way to tweak your videos is via the remix tool:
As you can see, there are three default remix options, along with a custom strength option.
Using it is simple. You just write in the text box what you’d like to change about the video. Then choose your strength and tap Remix.
It’s not a magic cure and I doubt it could do anything to help the Muay Thai video, but for something like the Bangkok video, you can get a decent final result:
The recut feature
The other option to improve a bad video requires that the bad video contains at least a second of something that you like about it. You can then use the Re-cut tool to clip out that one good part. Feed the part back into Sora, and have it extend the part into a longer clip.
This tool is also useful for creating your own extended clips that go beyond the time limit imposed by Sora.
For example, let’s say you have a good 5 or 10 second video, but you want to double or triple its length. You can do that by taking the last second or half-a-second of it. Then generate a new video with that tiny segment as the starting point. Once you have the original video and the extension, load them into another video editor (e.g., iMovie) and combine them for a longer clip.
Blending two videos together 🔁
Another interesting tool that Sora offers is called Blend. The name gives away what it does.
This tool also offers three default choices, along with a custom option. You can play around with it and get weirdly creative. Check out the result I got from blending my remixed Bangkok video with the other successful video I made earlier:
Final thoughts on Sora 💭
Overall, Sora was fun to play with. But even in the best case scenarios where you can get a decent video out of it, the only real-world application I can see it being used for is brief B-reel footage.
I mean what else are you going to do with a five to ten second video?
Even if you tried to string a story together by using the Re-cut feature as I described earlier, it would take you hours to get even a single minute of something usable.
At this point, I would argue that Sora is in its DALL E-1 stage. Although I never personally used DALL E-1, I do remember trying DALL E-2, and the way I felt about it is how I’m feeling about Sora.
It’s this kind of cool-but-not-quite-there-yet AI tool.
My personal guess is that OpenAI is probably using this initial rollout as a way of gathering massive amounts of new training data “from the wild.” They will then use this training data to improve Sora and hopefully in a few months it’ll be more than just a fun novelty. Maybe they’ll roll out access to more regions by then as well. Either way, I’ll be here for it.
Have you tried Sora yet? What did you think about it? Do you think it’s useful beyond just something to play around with? Let me know in the comments. I’ll see you there.
…
Don’t forget to join our crash course on speeding up your WordPress site. Learn more below:
