Moderation
While enjoying the generous holiday 2024 offer of unlimited Relaxed mode video generation in Sora, I was exploring the “Recent” feed for Text to video prompting ideas. Unfortunately I had to use the Report this video feature a lot. Since my original post, the moderation has improved quite a bit. So, don’t know why it was lax, but whatever was done to improve things, thank you, Sora team.
Mini-feedback & review
- Lots of wasted generations because of current limitations = wasting time and electricity. Sora can’t spell well at all
- The demos before the release clearly had more cycles to complete their generations, which if we had access to would avoid some but not all of the oh the person turned so the back of their head is going to morph into their face, as well as unrealistic physics. Please give us a way to allocate and spend more credits for more cycles for more accuracy, and some mid-tiers, please. Or relaxed mode access at very low use times.
- How about multimodal with audio, which might enable lip syncing to voices? That’s near impossible to do right now using Sora.
- Please More training data on physics and objects that are consistent failures, like (for example) aircraft. It’s seriously bad at rendering them as well as animating them and understanding their physics.
Thanks, and great job so far. But Please take everything you learn and make Sora 2.0 soon. And thank you.