I tried out OpenAI’s Sora, but I’m struggling with its learning curve

After months of waiting, it finally happened – OpenAI launched Sora, a video generator. Or, at least, it opens up access to the tool, only for the entire Internet to join in simultaneously, forcing Open artificial intelligence Put the brakes on account creation. Thanks to a little patience and determination, I’ve made it through the waitlist and now have the ability to produce almost anything I can think of within some well-defined limits.

However, with great power and responsibility comes something else: a huge learning curve. Although I love Sora and am impressed with its capabilities, I’m having a hard time identifying the perfect tips to get videos I’m happy with. I’m sure it’s just a matter of practice, but that’s what Sora and I experienced in those first few days together.

First, let’s talk about how Sora works – or at least how to access the powerful video generation tool. Although it comes from OpenAI, and you need to be a ChatGPT Plus or Pro member to start creating, you can’t access Sora through the main pipeline Chat GPT interface. Instead, you must go directly to the Sora website (SolaNet), where you’ll find a selection of clips that set the standard incredible High.

At least, they set a high standard in my mind. I scrolled through a few of them, looked at their tips, watched them run smoothly, and thought I could do the same thing. After all, my tips will run through the same DALL-E 3 adaptation as theirs, so they should look just as good, right? It’s not that easy. Sure, it’s easy enough to enter prompts, but it’s a little harder to figure out what Sora responds to best.

Before we get to the challenge, though, I should probably clarify some of Sora’s current limitations. Unlike Google’s Pixel Studio or other basic image generators, you can’t just sit down and run Sora on a whim — at least not with a $20-a-month ChatGPT Plus membership.

Instead, you’ll get 1,000 points, which you can use for movie generation if you want. Everything you change in a prompt, from aspect ratio to duration to resolution, will cost a certain number of points until you run out of points for the month. Fees for new videos range from 20 points to 2,000 points, you can view a helpful fee schedule here — I wish I could find it before spending 260 points in about 20 minutes. As a Plus member, you can only generate one video at a time, and the maximum resolution is 720p.

If you want to become a ChatGPT Pro member, the restrictions are much looser, but the price is much higher at $200 per month. For priority videos, you’ll get 10,000 points instead of 1,000, after which you’ll get unlimited video generation; they just take longer – OpenAI calls them “easy videos.” Pro members can also generate 5 videos at a time, upscale them to 1080p, and let them run for up to 20 seconds.

But unfortunately, no matter which level of ChatGPT you buy, Sora’s videos don’t have audio, so you’ll have to download the clips and sync the music or sound effects after settling on the visuals. OpenAI says Sora will eventually support audio, but that’s not yet available.

Now that you’ve covered the basics, generating the rest of your movie using Sora should be easy, right? Well, yes and no. Despite the input prompts, select Settings from the bottom menu and wait for the video to generate yes As simple as that is, it’s much harder to come up with something worthy of Sora’s ever-changing curated feed.

To share this month’s limited token cache, I contacted my colleague Mitja as soon as I visited Sora. He and I have been talking about how quickly we can get access to the platform, so I thought he might come up with some great ideas right off the bat that would be multi-generational. Turns out, his first idea was something I couldn’t have imagined: ten zebras in suits eating pesto ravioli and dancing to a Michael Jackson song in front of the Sydney Opera House. This may seem like a weird video to make, but if Sora can handle that much detail, it’s definitely the real deal.

After I laughed at the idea, I ran it on empty and waited for the results. Technically, much of the final product is correct. It placed a group of zebras in suits in front of the Sydney Opera House, all holding green plates. However, the number of zebras fluctuated between 8 and 12, there was no indication that it was a Michael Jackson song, and the pesto ravioli was definitely just a green plate – close, but not quite right. More worryingly, I increased the cost of the video to 100 tokens because I wanted the 10 second clip to show more dancing. This is not the case.

However, I later learned that Sora’s storyboarding tool is pretty much a must-have for anything involving complex motion. It lets you drag and drop clips along a five- or ten-second timeline, helping Sora break up the action and flow from one direction to another. So to get my zebra friends to draw more moves, I jumped into storyboarding and separated the dance and pesto ravioli into two separate moves, spaced out in five-second clips, which I then used ChatGPT to punch up My Description – Another built-in feature of Storyboards.

Again…it kind of works, but kind of doesn’t. Yes, I got zebras and they were in front of the Sydney Opera House, but they had given up dancing and when asked to eat some ravioli, they suddenly grew human hands to hold a fork. Sorry for the feature feed, but I guess I still have a long way to go.

I’ve also tried more natural cues, like macaroni penguins sliding off an iceberg into the ocean, and more fantastical ones, like a piece of toast with a Pixar face jumping out of the toaster, and the story is mostly the same. Sora does some parts of each prompt really well, but you have to describe your scene with just the right amount of detail. There were so many that Sora began to blend different elements. Too little and you’ll end up with a relatively bland finished product.

Yet, somehow, there’s more to Sora than I’ve ever been exposed to, especially when it comes to edit. Video Generator also features the ability to recut, remix, and remix clips to expand creative ideas, connect one video to another, or cut out elements that don’t work well. But again, I still want to nail down a video that looks good the first time.

Overall, it’s fair to say that my first few days with Sora were a mixed bag. Is the video generator perfect yet? No, but I can’t blame this entirely on OpenAI. This was my first attempt at generating a video purely based on text, so I wasn’t surprised that I struggled to get the level of detail right. So far, I’ve given Sora too much information and too little information, which means that identifying the right hints is just around the corner.

But more importantly, I was impressed by Sora’s commitment. I can create videos as ChatGPT Plus member Just take a moment to imagine it, and I imagine as the models continue to train, they will get faster. I’m not quite sure if I’ll use any of the quick clips Sora made – many of which still suffer from weird artifacts, such as the appearance of a human arm on my zebra – but the clips that did appear in Sora’s featured collection give me hope , it’s just a matter of learning how to go for the right elements.

Beyond that, I wouldn’t be surprised if the way OpenAI handles prompts and creation is also open. Now, when you’ve exhausted 1,000 points as a ChatGPT Plus member, that’s it — you can’t purchase more points until the end of the billing cycle. Likewise, there’s no way to roll over unused points from one month to the next, so you have to find the right balance of spending and saving to get through the month.

If it were up to me, I’d definitely want to get some of the stupid points I spent back, but that’s not an option. Instead, I call it a learning cost, I just have to spend a little more time fine-tuning the prompt before sending it to Sora. Maybe one day I’ll come up with something worth recommending.