SORA 2 EARLY ACCESS EXPERIMENTS – LANDSCAPE

(compression does not represent the raw footage of Sora 2. There’s also a glitch in my editing software that is creating audio imperfections. Will update as I can!)

▫️Project: Stress test Sora 2 while it is under Early Access and Invite Only.

▫️Experiment: How comprehensively does Sora 2 generate a commercially viable scene from a single sentence prompt.

▫️Tools: Sora 2 (Text to Video), DaVinci (editing). 

▫️Vision: Create a reel of visually and emotionally diverse scenes that evaluate the platform’s ability to act as a full-stop production house. 60 second video, with six 10-second clips. Landscape formatting only (portrait is in another experiment).

▫️Expectations: Sora 2’s Portrait mode generations impressed me, and I quickly discovered you can flip to landscape formats as well. So I decided to dedicate a compilation to each specific format, knowing that Ai slavishly builds for the space you define. My hope was that Landscape’s “theatrical” formatting would encourage better cropping and by extension, better scenework – as opposed to Portrait mode, which naturally lends itself to found footage.   

▫️What actually happened: My initial awe at Sora 2 dimmed quickly when Landscape formatting revealed the ugly truth: Ai is still a big roulette wheel spin. You really feel the wasted clips under their 30 gen limit.  I’d wager for every 3 generations I made, 1 was viable. While that’s better than Midjourney’s ratio for video, the stunning fidelity of Sora 2 makes it feel like there’s far more to mess up. If the devil is in the details, Sora 2 just gives the Lord of Lies more places to hide and wait this all out. 

▫️What I didn’t expect: The voice over work, while initially charming, is impossible to control – certainly within a single sentence prompt. While I appreciate the expressiveness and variety; power without control ain’t very powerful. It’s strange: Sora 2 will sometimes try to speed through a lot of dialog in a short time, but won’t encapsulate it in the 10 second clip. You get the distinct impression it’s generating for far more runway and *then* gets cut to the 10 second mark. I have no doubt this will improve but for now, it’s a double-edged sword. And no, I haven’t tried to prompt to dialog at all since I find the evocative voice acting intriguing. 

▫️Results: Portrait mode was my first run with Sora 2 and I think the awed honeymoon period is over with Landscape mode. Between wasted clips, madcap voice overs (and I’ll include music in that mix) and the 10 second hard stop – it’s a stunning tech demo that leaves a lot of questions for its commercial viability. Yes, you could make a full-blown commercial, but how would you tweak every little nuance? Sora 2 is like a rocketship: incredibly powerful, but you’re just along for the ride. 

🔴If You’re Anti Ai: Sora 2 is in Early Access with a waiting list, so you have time before it proliferates. Its enterprise use won’t be available for some time, so you can make your case that integrating it into any pipeline is premature. It’s also too short in its output right now to make a judgement call on day-to-day operations, so that’s another point for you to bring up. Finally, Sora 2 is pretty draconian in what you can generate, so you can still market yourself as much more accommodating than their “guardrails”.

🔵If You’re Pro Ai: Get on the Waiting List today. In the meantime, take a look at the gens people are making and get a sense of the differences in Portrait versus Landscape formats. As mentioned earlier, Portrait is good for found footage but the Ai will try its best to make your Prompt work in an traditionally awkward space. Landscape, however, is likely where most of the enterprise use will happen. No matter how much the tech changes, these two formats (minor nod to square) will be your bread and butter.