Review of OpenAI’s Sora on its first day: Unpredictable. Expensive. Overhyped and underwhelming. Rated 2 out of 5 stars.
You may think this video that I generated with Sora is pretty and cool but it did not follow my prompt. I wanted “watercolor paints flowing down a canvas to form roses among green leaves” but instead I got a hand painting the roses. This video was the better of two variations. The other had a very unnatural painting motion by the unwanted hand.
This video was only my 8th prompt today at the lowest 480p resolution, and I had already used HALF my Sora credits for the month. OpenAI thinks that you will subscribe to its $200/month ChatGPT Pro plan instead of the $20/month ChatGPT Plus (that I have) so you can get unlimited 1080p video generations out of Sora. NOPE.
In my other attempts, I tried generating animals and physical objects. A running horse looked like it was bouncing on the same spot. Rotating cubes suddenly took on a ribbed texture that I didn’t ask for. I kept my prompts simple but it didn’t seem to reduce the weirdness
There will be many people defending Sora’s performance today as “this is the worst AI will get” but honestly, it’s been oversold by OpenAI’s marketing team since the unveil in Feb. I would say that most of the videos I generated today were unusable and not fit for posting, given their weird animation or sometimes, complete lack of animation (Sora just gave up animating an image I gave it).
Yes, yes, I know Sora will improve like all Gen AI products. I *was* a big advocate of Sora, having showed off the marketing video samples in my AI workshops since Feb when Sora was first announced. Now, most people will find it hard to justify a paid ChatGPT account just to generate a dozen or so unpredictable videos a month.
Let’s see how AI video competitors Runway and Kling respond to Sora’s debut. I’m also waiting for Adobe to give me access to its text-to-video generator.
Sorry, Sora, I wanted to love you more.
