After transforming the world with text-based generative AI, OpenAI is back and now, it can generate videos. OpenAI introduced Sora, its AI text-to-video diffusion model, earlier on Thursday that can generate photo-realistic videos based on user prompts. It is currently available for red teams and a group of experts and visual designers who can use it through and through to provide insights and valuable feedback before the AI model is made available to the public.
OpenAI Sora set to disrupt text-to-video AI market
OpenAI disrupted the tech world with its ChatGPT which currently has 100 million users. It started with text-based prompts and generative AI and has introduced voice prompts and image prompts, the latest ChatGPT 4.0 is connected to the internet to furnish updated data and so on. With Sora, OpenAI is set to disrupt AI video generation capabilities.
You might have seen video-generating models on the internet, however, these are limited to a few seconds or may not have promising results. Google is working on a text-to-video model, Meta already has one but for short videos. However, OpenAI Sora can create AI videos based on text up to a minute long and can perfectly emulate realistic graphics and complex settings such as a couple of people walking through on the sidewalk of a busy road that would require immense computing and processing.
According to OpenAI, Sora can contemplate complex scenes involving specific types of motion, multiple characters, vibrant emotions, and attention to details in the background gracefully. The AI analyses user prompts and understands what the user wants and how it will exist in the physical world to create realistic videos based on text.
The AI still requires fine-tuning given the early stage. For instance, it might not understand the cause and effect of certain scenarios such as no cookie eating marks when someone takes a bite off the cookie or it could be specific to the camera trajectory to follow or others. However, we can confidently say that just like ChatGPT which took a trove of data to train upon and furnish promising results, Sora will get to that level soon.
OpenAI is fortifying the safety nets on Sora
When it comes to safety, OpenAI will use safety nets to prevent users from generating videos based on hateful sentiments, sexual, extreme violence, likeliness to celebrities, or using IPs reserved by others. Additionally, it will use sophisticated image classifiers to scan through each frame of videos generated to keep the results as per usage policies.
Following the C2PA metadata guidelines, OpenAI will watermark videos generated using Sora (its text-to-video diffusion model) to identify original versus AI-generated videos. This should help prevent the spread of AI videos (often marked as legit or real) reducing the virality of unwanted content.
You can follow Smartprix on Twitter, Facebook, Instagram, and Google News. Visit smartprix.com for the most recent news, reviews, and tech guides.