171–180 of 916 posts
I think feeling like you need to use that in marketing copy is a pretty good clue in itself both that its not, and that you don’t believe it is so much as desperately wish it would be.
As usual: impressive until you look close. Just freeze the frame and you see all the typical slop errors: pretty much any kind of writing is a garbled mess (look at the camera in the beginning). The horn of the unicorn sits on the bridle. The buttons on Sam's circus uniform hover in the air. There are candleholders with somehow candles inside as well as on top. The miniature instruments often make no sense. The condu…
Anyone, literally anyone, can use it (eventually) to generate incredible scenes. Imagine the person who comes up with a short film about an epic battle between griffins and aliens...Or a simple story of a boy walking in the woods with their dog...Or a story of a first kiss. Previously people were limited to what they had at hand. They couldn't produce a video because it was too costly. Now they can craft a video to meet their vision.
I do find it exciting.
The most interesting thing by far is the ability to include video clips of people and products as a part of the prompt and then create a realistic video with that metadata. On the technical side, I'm guessing they've just trained the model to conditionally generate videos based on predetermined characters -- it's likely more of a data innovation than anything architectural. However, as a user, the feature is very coo…
This is something I would not like to see, I prefer product videos to be real, I am taking a risk with my money. If the product has hallucinated or unrealistic depiction it would be a kind of fraud.
Earlier quoted context omitted.
I actually wonder if this will kill off the social apps and the bragging that happens. It will be flooded by people faking themselves doing the unimaginable.
This is also my thesis. The internet is going to be saturated with AI slop indiscernible from real content. Once it reaches a tipping point, there will no longer be much of a reason to consume the content at all. I think social networks that can authenticate video/photo/text content as human-created will be a major trend in a few years.
Unless there some fundamental, technical way to distinguish the two, I wonder who would win?
One use that occurred to me is that fans will be able to "fix" some movies that dropped the ball. For example, I saw a lot of people criticizing "Wish" (2023, Disney) for being a good movie in the first half, and totally dropping the ball in the last half. I haven't seen it yet, but I'm wondering if fans will be able to evolve the source material in the future to get the best possible version of it. Maybe we will eve…
Much more mundane, but useful!
VFX artists are definitely feeling the AGI / considering other career paths today.
> new AI feature/model comes out
> "it's going to replace people in this field! they better start looking for a new job!!!"
why is this a good thing?
“With Sora 2, we are jumping straight to what we think may be the GPT‑3.5 moment for video.” I think feeling like you need to use that in marketing copy is a pretty good clue in itself both that its not, and that you don’t believe it is so much as desperately wish it would be.
Sora 2 itself looks and sounds a little poorer than Google Veo 3. (Which is itself not currently ranked as the top video model. The Chinese models are dominating.)
I think Google, with their massive YouTube data set, is ultimately going to win this game. They have all the data and infrastructure in the world to build best-in-class video models, and they're just getting started.
The social battle will be something completely different, though. And that's something that I think OpenAI stands a good chance at winning.
Edit: Most companies that are confident of their image or video models stealthily launch it on the Model Arena a week ahead of the public model release. OpenAI did not arrange to do that for Sora 2.
Nano Banana, Seedream/Seedance, Kling, and several other models have followed this pattern of "stealth ELO ranking, then reveal pole position".
https://artificialanalysis.ai/text-to-video/arena?tab=leader...
The fact that this model is about "friends" and "social" implies that this is an underpowered model. You probably saw a cherry picked highlight reel with a large VRAM context, but the actual consumer product will be engineered for efficiency. Built to sustain a high volume of cheap generations, not expensive high quality ones. A product built to face off against Meta. That model compete on the basis of putting you into videos with Pikachu, Mario, and Goku.
it doesn't spark optimism or joy about the future of engaging with the internet & content which was already at a low point.
old is gold, even more so
The most interesting thing by far is the ability to include video clips of people and products as a part of the prompt and then create a realistic video with that metadata. On the technical side, I'm guessing they've just trained the model to conditionally generate videos based on predetermined characters -- it's likely more of a data innovation than anything architectural. However, as a user, the feature is very coo…
> the ability to include video clips of people and products as a part of the prompt and then create a realistic video with that This is something I would not like to see, I prefer product videos to be real, I am taking a risk with my money. If the product has hallucinated or unrealistic depiction it would be a kind of fraud.