We’ve just wrapped up our first night of testing OpenAI’s Sora model. A few hours isn't enough to form a definitive opinion, but since we’ve recently been deep in testing various video generation tools at bards.ai for different projects, I thought I’d share some early impressions - especially in comparison to the current competition.

Here is short comparison video we made: https://yoube.com/shorts/ub89TnA_JXY

Overall take? Very positive - but to be honest, Sora (at least the Turbo version that’s currently available) doesn’t feel revolutionary.

In terms of generative video quality, Sora is most comparable to Kling AI. Is it better? Worse? Honestly, it depends on the prompt. To get a truly objective answer, we’ll need to wait for an ELO leaderboard for video models 😁 My hunch? These two will land pretty close together.

Where Sora really stands out is as a product, especially when compared to Kling. A few key examples:

Length limitations: Kling 1.5 currently caps videos at 10 seconds. Sora allows up to 20 seconds. Arbitrary numbers, sure - but having 20-second clips makes it significantly easier to stitch together something coherent.

User experience: Sora is designed with consumers in mind, with tools focused on producing a finished video. Kling, by contrast, is more of a raw video generator - powerful, but bare-bones.

As for the rest of the competition - LumaLabs and Runway - they tend to generate a different type of video altogether. Their outputs are often more artistic, with a distinct separation between background movement and character motion (which is often quite limited). Whether that’s good or bad depends on the use case, but it’s certainly different.

Between the two, we usually prefer LumaLabs, mainly due to fewer visual artifacts.

To give you a sense of the current landscape, I’m sharing two video generations from each model using the exact same prompt. No cherry-picking - just the first two results we got.

In this example, Runway clearly lags behind the others. That lines up with our general impression at bards.ai, though the gap is usually less dramatic - probably just an unlucky generation here. Still, I promised no cherry-picking, so what you see is what you get ;)