OpenAI is finally here Soraits advanced text-to-video generator, made available to the public. Like you can generate thousands of words by sending a quick reminder above ChatGPTYou can now create videos by sending a prompt on Sora. But its functionality goes beyond that—you can also upload images to bring them to life. Imagine sending an old photo of your great-grandfather and watching him walk, run, or perform other actions. That’s the level of innovation we’re talking about here. That said, you may be wondering: How is Sora different from LLM (Large Language Model) like GPT-4 that supports ChatGPT, how is the technology behind Sora? What is it trained for? And most importantly, can Sora be accessed by everyone? We answer them all below—read on.
Also read: iPhone 18 Pro costs a lot more, the main reason for the price increase is…
How is Sora different from large text-based language models?
First, Sora is a text-to-video AI model, while GPT is a Large Language Model. Despite their differences, the input they accept is somewhat similar, as GPT-4 is multimodal—it can handle text, images, and even video as input.
For example, with Sora, you can create videos of mountain slopes by sending detailed reminders. You can specify whether the mountains should be covered with snow, whether the sun should shine or not, and other details. Furthermore, you can breathe life into an existing image by sending it to Sora. In essence, Sora processes text, images or video as input and produces video as output.
In contrast, the GPT model only produces text output, even if you send text or image input. This difference in output sets the two apart.
You can ask: ChatGPT or Google Gemini can create images–Aren’t they provided by GPT-4 and Gemini respectively? The answer is no. They rely on models like DALL-E 3 (OpenAI) and Imagen 3 (Google) to generate text into images.
Sora also has the ability to scale video, moving forward or backward in time, adding another dimension to its capabilities.
Also read: iOS 18.2 released this week: iPhone users will get powerful AI features on…
How was Sora trained?
OpenAI claims that Sora was trained on videos and images of different lengths, resolutions, and aspect ratios. The company claims that it uses the Transformer architecture, which handles spatio-temporal patches of video and image latent code.
From a technical perspective, there is a significant difference in approach. While text-to-video models like Sora are trained differently, large language models like GPT-4o or other AI models from companies like Meta’s Llama are often trained on the same things. What we call tokens.
However, instead of using tokens, OpenAI uses a method called visual patching to train Sora. It essentially divides the video into patches by compressing them into a lower dimensional latent space. This representation is then decomposed into space-time patches for further processing.
Who is Sora for?
Currently, free users cannot access Sora. This means you’ll need to purchase a subscription — an OpenAI Plus subscription or an OpenAI Pro subscription.
Plus subscription comes at a price $2,000 in India, giving access to 50 generations of Sora videos every month. If you choose the Pro account which costs $200, you can create up to 500 generations of fast videos. However, choosing a higher resolution will reduce the number of generations available. If you’re patient and willing to use the slower creation mode, you can enjoy unlimited generations of videos.
It’s important to note that there is a limit to resolution and videos can only be up to 20 seconds long. They are available in wide, portrait, or square aspect ratios.
For specifics, OpenAI Plus (or ChatGPT Plus) users can create videos at 480p resolution, with a limit of 50 videos. Although 720p can be produced, the amount of video captured will be less.
That said, due to overwhelming demand, OpenAI is currently not allowing new registrations for Sora. OpenAI CEO Sam Altman confirmed that registrations have been temporarily paused but will resume once demand stabilizes. He assured users that OpenAI is doing its best to resolve these issues as quickly as possible. However, it may take some time before everyone has access to Sora. If you don’t get early access it’s because of high demand.
Also read: iPhone 17 Air could be 2mm thinner than iPhone 16 Pro: See more details here
