OpenAI SORA : A Text-to-Video AI After ChatGPT
We already have functioning robots. We have AI tools that can write or present any information. Now, we have OpenAI SORA, which has brought AI-generated videos into existence. So, should we expect robots to take over soon because of so much AI development? Hopefully not, but Sam Altman has created another super AI tool, so let’s talk about OpenAI’s new product, SORA.
Understanding OpenAI SORA Technology
OpenAI has previously introduced us to ChatGPT, which has been of great help to everyone. This time, it’s a step ahead as now Video generation is possible with OpenAI SORA.
SORA, a Japanese word that means SKY, is the world’s first-ever AI technology that makes realistic videos. These videos are up to 1 minute long, and all the images seem real. SORA can maintain cohesion between frames, and images can be rendered in different aspect ratios. For SORA, Sam has collaborated with various graphic designers, visual artists, and designers to have feedback on the performance of SORA over time.
This news came on 15th February when Sam posted about this on his personal X account. Not only this, he live-streamed the working and functioning of SORA in front of thousands of watchers on the platform as well. He drafted on-camera scenarios like ‘two dogs doing podcasts on the top of the mountain, flying squirrel on dragon’ to showcase the AI power of SORA.
Technical Features of SORA
SORA can generate 275 videos of up to 1 minute in length, either by a text prompt or an image for reference, it will fill it with motion.
Open models like Video Diffusion and a private product called PIKA already exist, but SORA has taken over everything with its better technology. It is like the diffusion model DALL-E, which starts with generating random noises and then gradually updating those noises into coherent images.
A single image is of tonnes of data, like 1000X1000 pixels in three color panels, making it to millions of data points. The videos created by SORA are 1 minute long, and with multiple time frames, the data points will be humongous. These data are called patches and are similar to the tokens used as data units in ChatGPT.
SORA doesn’t tokenize texts. Instead, it views them as visual patches that are small compressed chunks of images that capture the visualization and movement in time of by frames.
What’s unique about SORA is that it doesn’t crop the training data and outputs for specific time and resolution like other training models. Instead, it trains data on its native resolution and output variable resolutions.
OpenAI has claimed that SORA can generate complex scenes that can include multiple characters and any kind of motions while providing details on background and subject. And the last, it can extend any video by filling in missing frames.
Limitations of SORA
SORA is a great technology, and users must be waiting to use it. But that’s not possible, at least for now. SORA isn’t accessible to people as this can get misused. The company has limited it to some specific people only for research purposes. In a different circumstance, if it gets public, it will have C2PA metadata. It is a surveillance apparatus that will record all the entries, like where the content came from and all modifications.
OpenAI SORA does have some massive flaws. The videos created are of good quality, but it does have a distinct AI touch that can differentiate them from human-created videos. Aside from that, it hasn’t perfectly modeled physics and humanoid interactions.
Also, SORA is not cost-effective. It requires massive computational power. Sam Altman asked its audience for 7 Trillion in funding to buy some GPUS to get more computational power. Which later received many trolls, including from NVIDIA CEO Jensen Huang.
Conclusion
OpenAI SORA has taken over the market as everyone wants to get their hands on it. It has introduced many useful features that save time while creating and editing videos. The OpenAI team has discussed bringing more updates and working on the limitation soon. It’s time to see what else is left for us to see in the development of AI after this. Read about How OpenAI value soared after SORA.
The post OpenAI SORA : A Text-to-Video AI After ChatGPT appeared first on CoinGape.
Read More
OpenAI SORA : A Text-to-Video AI After ChatGPT
We already have functioning robots. We have AI tools that can write or present any information. Now, we have OpenAI SORA, which has brought AI-generated videos into existence. So, should we expect robots to take over soon because of so much AI development? Hopefully not, but Sam Altman has created another super AI tool, so let’s talk about OpenAI’s new product, SORA.
Understanding OpenAI SORA Technology
OpenAI has previously introduced us to ChatGPT, which has been of great help to everyone. This time, it’s a step ahead as now Video generation is possible with OpenAI SORA.
SORA, a Japanese word that means SKY, is the world’s first-ever AI technology that makes realistic videos. These videos are up to 1 minute long, and all the images seem real. SORA can maintain cohesion between frames, and images can be rendered in different aspect ratios. For SORA, Sam has collaborated with various graphic designers, visual artists, and designers to have feedback on the performance of SORA over time.
This news came on 15th February when Sam posted about this on his personal X account. Not only this, he live-streamed the working and functioning of SORA in front of thousands of watchers on the platform as well. He drafted on-camera scenarios like ‘two dogs doing podcasts on the top of the mountain, flying squirrel on dragon’ to showcase the AI power of SORA.
Technical Features of SORA
SORA can generate 275 videos of up to 1 minute in length, either by a text prompt or an image for reference, it will fill it with motion.
Open models like Video Diffusion and a private product called PIKA already exist, but SORA has taken over everything with its better technology. It is like the diffusion model DALL-E, which starts with generating random noises and then gradually updating those noises into coherent images.
A single image is of tonnes of data, like 1000X1000 pixels in three color panels, making it to millions of data points. The videos created by SORA are 1 minute long, and with multiple time frames, the data points will be humongous. These data are called patches and are similar to the tokens used as data units in ChatGPT.
SORA doesn’t tokenize texts. Instead, it views them as visual patches that are small compressed chunks of images that capture the visualization and movement in time of by frames.
What’s unique about SORA is that it doesn’t crop the training data and outputs for specific time and resolution like other training models. Instead, it trains data on its native resolution and output variable resolutions.
OpenAI has claimed that SORA can generate complex scenes that can include multiple characters and any kind of motions while providing details on background and subject. And the last, it can extend any video by filling in missing frames.
Limitations of SORA
SORA is a great technology, and users must be waiting to use it. But that’s not possible, at least for now. SORA isn’t accessible to people as this can get misused. The company has limited it to some specific people only for research purposes. In a different circumstance, if it gets public, it will have C2PA metadata. It is a surveillance apparatus that will record all the entries, like where the content came from and all modifications.
OpenAI SORA does have some massive flaws. The videos created are of good quality, but it does have a distinct AI touch that can differentiate them from human-created videos. Aside from that, it hasn’t perfectly modeled physics and humanoid interactions.
Also, SORA is not cost-effective. It requires massive computational power. Sam Altman asked its audience for 7 Trillion in funding to buy some GPUS to get more computational power. Which later received many trolls, including from NVIDIA CEO Jensen Huang.
Conclusion
OpenAI SORA has taken over the market as everyone wants to get their hands on it. It has introduced many useful features that save time while creating and editing videos. The OpenAI team has discussed bringing more updates and working on the limitation soon. It’s time to see what else is left for us to see in the development of AI after this. Read about How OpenAI value soared after SORA.
The post OpenAI SORA : A Text-to-Video AI After ChatGPT appeared first on CoinGape.
Read More