The videos certainly are hilarious! Don't forget that the released implementation is "Sora Turbo" which implies that it's a small (thus fast) version of the model, not the largest, fully trained version. Quite a few people have reported that this version does much better when given lengthy, detailed prompts. Hopefully when they release the full version it will be able to generate more realistic videos when given short, ambiguous prompts.
You literally asked for synthetic clouds and then complained they weren't realistic... C'mon now - you know these models aren't mind readers and "words have meaning". Sora is amazing, jaw dropping tech... but we're at about GPT-3.5 levels of usefulness tho.
I just think it's funny how it goes against our expectations! True, they will improve and I asked for “synthetic” clouds though that’s not exactly what I envisioned when I pictured them
I think the prompts you used in the video only targeted things AI video generators generally struggle with. These AI models typically have difficulty with most things involving physics (for example, the Quidditch and the basketball) or complex body movements. Also, I have never seen an AI model successfully create a cross-section of the human brain, as that is, at least at the current stage, an extraordinarily difficult prompt. However, they’re really good at creating cartoons and drone shots where physical accuracy doesn’t matter as much. So maybe try that. But I agree Sora isn’t as big of a leap as i would have hoped.
Lmao. Nice, try using 1080 or at least 720p 16:9 generations, it's a lot more compute and credits but it does tend to give more realistic physics-looking generations.
It feels like a novelty or toy to me, not a tool like o1. If I'm watching video I want it to be coherant - Sora videos are chock full of elements that break the suspension of disbelief because it's scattered with nonsensical elements in the videos that make me know right away it's AI. It's hard to watch, even at best it ends up looking video game-like to me, at worst it's complete nonsense.
Please include the video understanding capabilities and capabilities try other than human being audio understanding capabilities and image recognition abilities
3:00 - The prompt says it's modelling the part of the brain which is part of the visual system. Not the entire brain. I'm not smart enough to know whether it's correct or not. But what you're mentioning as commentary doesn't seem to match the prompt given to Sora.
My first try at posting got auto-deleted, I think, because I included a link to one of my Sora-generated videos. So I'll try again. It clearly has no knowledge of how objects move and interact. I've given it pictures of cars and the resulting videos have hilariously unrealistic motion. And it definitely has big trouble with legs. But I had it generate an animated title slide for a video and it looked very good. I did get it to generate a video showing clouds, from above, as seen from the window of a passenger plane and it was decent. Way better clouds than yours. It's a lot of fun, and it'll only get better from here. I think they did a great job on the UI.
It's impossible for this technology to generate high-quality, realistic images with the current hardware. I believe the released version of Sora might not be the complete version (perhaps with fewer parameters). Even with that, OpenAI seems to have struggled a lot to make it available to the public in a way that provides an acceptable experience. Maybe in 5 to 10 years, we'll see much better outputs from this kind of technology. Result are so funny though 😂
Maybe you have to prompt Sora a bit differently than ChatGPT since Sora doesn't seem as good at inferring things from language. In my limited experiments with Sora, the more detailed and specific the prompt, the less weirdness there is in the result. If you leave too much to Sora's imagination, you get an AI fever dream.
AI is just a baby or a very young child, stop being so mean to it! 🤣 AI will grow exponentially and then we will not be laughing anymore! It might start laughing at us! 🤔
I have a different Percpective on it i think they are just nerfing thier sora modal that is available for public so to avoid misuse i think ai is atleast that developed that it can generate video perfectluy enough
@@gonzalezm244 It is both spectacularly brilliant at many tasks and completely idiotic at many others. Therefore not to be relied upon without in depth human supervision.
The videos certainly are hilarious! Don't forget that the released implementation is "Sora Turbo" which implies that it's a small (thus fast) version of the model, not the largest, fully trained version. Quite a few people have reported that this version does much better when given lengthy, detailed prompts. Hopefully when they release the full version it will be able to generate more realistic videos when given short, ambiguous prompts.
You literally asked for synthetic clouds and then complained they weren't realistic... C'mon now - you know these models aren't mind readers and "words have meaning".
Sora is amazing, jaw dropping tech... but we're at about GPT-3.5 levels of usefulness tho.
I just think it's funny how it goes against our expectations! True, they will improve and I asked for “synthetic” clouds though that’s not exactly what I envisioned when I pictured them
I think the prompts you used in the video only targeted things AI video generators generally struggle with. These AI models typically have difficulty with most things involving physics (for example, the Quidditch and the basketball) or complex body movements. Also, I have never seen an AI model successfully create a cross-section of the human brain, as that is, at least at the current stage, an extraordinarily difficult prompt. However, they’re really good at creating cartoons and drone shots where physical accuracy doesn’t matter as much. So maybe try that. But I agree Sora isn’t as big of a leap as i would have hoped.
lmao AI gen content has gone full circle to being goofy again with videos i love it. its like dalle1 was for images
It's because you used the term "synthetic"...!
and the cerebellum isn't part of the visual system...
Lmao. Nice, try using 1080 or at least 720p 16:9 generations, it's a lot more compute and credits but it does tend to give more realistic physics-looking generations.
It feels like a novelty or toy to me, not a tool like o1. If I'm watching video I want it to be coherant - Sora videos are chock full of elements that break the suspension of disbelief because it's scattered with nonsensical elements in the videos that make me know right away it's AI. It's hard to watch, even at best it ends up looking video game-like to me, at worst it's complete nonsense.
Make a video on gemini 2.0 flash testing o1 like question
Only good at creativity
That's a great idea! I'll add it to the list!
Please include the video understanding capabilities and capabilities try other than human being audio understanding capabilities and image recognition abilities
You used the word synthetic in the prompt. Wouldn't this necessarily give something that looks...synthetic?
3:00 - The prompt says it's modelling the part of the brain which is part of the visual system. Not the entire brain. I'm not smart enough to know whether it's correct or not. But what you're mentioning as commentary doesn't seem to match the prompt given to Sora.
My first try at posting got auto-deleted, I think, because I included a link to one of my Sora-generated videos. So I'll try again.
It clearly has no knowledge of how objects move and interact. I've given it pictures of cars and the resulting videos have hilariously unrealistic motion. And it definitely has big trouble with legs. But I had it generate an animated title slide for a video and it looked very good. I did get it to generate a video showing clouds, from above, as seen from the window of a passenger plane and it was decent. Way better clouds than yours.
It's a lot of fun, and it'll only get better from here. I think they did a great job on the UI.
Yeah, but i can have fun for 5$, why spend 200 on fun
It's impossible for this technology to generate high-quality, realistic images with the current hardware. I believe the released version of Sora might not be the complete version (perhaps with fewer parameters). Even with that, OpenAI seems to have struggled a lot to make it available to the public in a way that provides an acceptable experience. Maybe in 5 to 10 years, we'll see much better outputs from this kind of technology.
Result are so funny though 😂
Sora is good to create memes
Agreed!
Looks exactly like my dreams/nightmares... 😅😂
Maybe you have to prompt Sora a bit differently than ChatGPT since Sora doesn't seem as good at inferring things from language. In my limited experiments with Sora, the more detailed and specific the prompt, the less weirdness there is in the result. If you leave too much to Sora's imagination, you get an AI fever dream.
dont laugh, maybe ai will spare you when it takes over
ty
lol i laughed so hard too
AI is just a baby or a very young child, stop being so mean to it! 🤣 AI will grow exponentially and then we will not be laughing anymore! It might start laughing at us! 🤔
I have a different Percpective on it i think they are just nerfing thier sora modal that is available for public so to avoid misuse i think ai is atleast that developed that it can generate video perfectluy enough
And some will pay 20 to 200$/mth for AI with this junk :-)
ppl waste money on more frivolous things all the time why not llm services
O1 is much better at university work than 4o to be fair
@@gonzalezm244 It is both spectacularly brilliant at many tasks and completely idiotic at many others. Therefore not to be relied upon without in depth human supervision.
@@citogrid indeed, o1 can be very dumb. But 4o is substantially dumber to the point that it’s often unusable for more math heavy applications.