Sora for dummies: 101 on OpenAI's new text to video AI model

Artificial intelligence (AI) is developing at breakneck speed and this weekend, OpenAI delivered one of the biggest updates to the system that we have seen in a while. Sora is OpenAI's newest AI model that can create realistic and imaginative scenes simply with text instructions. 

With Sora, industry professionals will now be able to create realistic and complex videos all without leaving their seat.

This is more important than ever before especially because consumers today watch more videos and the demand for short-form content has rapidly increased, with 66% finding the type of content to be the most engaging, according to a report done by Munch, an AI-powered automation platform for social media.

Don't miss: Why OpenAI's comms strategy was a tragic fail amidst the Sam Altman ousting saga

According to the report, video content is no longer an option, but a necessity for business and brands aiming for success, with 42% of businesses preferring Instagram and 26% preferring Facebook to post such videos. TikTok does not rank among the top three platform choices for marketers. 

With the importance of short-form video content in marketing efforts, here's a breakdown of exactly what you need to know about Sora and how it can help industry professionals in the space. 

What is Sora?

Sora is OpenAI's solution to getting AI to understand and simulate the physical world in motion, with the goal of training models that help people solve problems that require real-world interaction, it said in a statement. 

As such, Sora is a text-to-video model that can generate videos up to a minute long while maintaining visual quality and adherence to the user’s prompt.

Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background. The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.

The model has a deep understanding of language which then enables it to accurately interpret prompts and generate compelling characters that express vibrant emotions. Sora can also create multiple shots within a single generated video that accurately depict characters and visual style.

"Specifically, we train text-conditional diffusion models jointly on videos and images of variable durations, resolutions and aspect ratios. We leverage a transformer architecture that operates on spacetime patches of video and image latent codes," said OpenAI.

How does it work exactly?

This part is a bit technical but according to OpenAI, it takes inspiration from large language models which acquire generalist capabilities by training on internet-scale data.

"The success of the LLM paradigm is enabled in part by the use of tokens that elegantly unify diverse modalities of text—code, math and various natural languages. In this work, we consider how generative models of visual data can inherit such benefits," it said. 

OpenAI explained in its technical report that where LLMs have text tokens, Sora has visual patches. Patches have previously been shown to be an effective representation for models of visual data.

"We find that patches are a highly scalable and effective representation for training generative models on diverse types of videos and images," it said. 

Sora then is essentially a diffusion model, which generates a video by starting off with one that looks like static noise and gradually transforms it by removing the noise over many steps.

It is, as a result, capable of generating entire videos all at once or extending generated videos to make them longer.

The model also builds on past research in DALL·E and GPT models. It uses the recaptioning technique from DALL·E 3, which involves generating highly descriptive captions for the visual training data. As a result, the model is able to follow the user’s text instructions in the generated video more faithfully.

In addition to being able to generate a video solely from text instructions, the model is able to take an existing still image and generate a video from it, animating the image’s contents with accuracy and attention to small detail.

What are some of its weaknesses?

As with all AI models, there are weaknesses, bias and misinformation that can sometimes arise. Sora is no exception as OpenAI admits. 

Currently, Sora may struggle with accurately simulating the physics of a complex scene and may not understand specific instances of cause and effect. For example, a person might take a bite out of a cookie, but afterward, the cookie may not have a bite mark, said OpenAI. 

The model may also confuse spatial details of a prompt, for example, mixing up left and right, and may struggle with precise descriptions of events that take place over time, such as following a specific camera trajectory.

Ahead of its public launch, OpenAI said that it will be working with domain experts in areas such as misinformation, hateful content, and bias — who will be adversarially testing the model.

"We’re also building tools to help detect misleading content such as a detection classifier that can tell when a video was generated by Sora. We plan to include C2PA metadata in the future if we deploy the model in an OpenAI product," it said. It added that it also will utalise its existing safety methods that it has already built for products that use DALL-E 3. 

For example, once in an OpenAI product, its text classifier will check and reject text input prompts that are in violation of its usage policies. These include those that request extreme violence, sexual content, hateful imagery, celebrity likeness, or the IP of others.

"We’ve also developed robust image classifiers that are used to review the frames of every video generated to help ensure that it adheres to our usage policies, before it’s shown to the user," it said. 

OpenAI will also engage policymakers, educators and artists around the world to understand their concerns and to identify positive use cases for this new technology.

Adopting the technology in marketing

With all that said, the main question really still lies in how marketers and industry professionals can adopt the technology in their day-to-day work and according to industry professionals MARKETING-INTERACTIVE spoke to, it has the potential to be "stunning".

According to Pramodh Rai, co-founder at Cyber Sierra, the capabilities, and low barrier to entry by Sora means that consumers have a "very high chance" of igniting experimentation in creative teams that include marketers and advertisers.

"Content creation is now revolutionised in very exciting ways. Our ability to prototype rapidly and produce high quality videos as well as dynamic visuals significantly reduces time to market and resources required for traditional ad campaigns," said Rai. He added that routine editing tasks and content tailored to different platforms as well as audiences can be automated, which frees creative teams to focus on strategic and innovative aspects of their campaigns. 

"As the line between reality and AI is blurring thanks to advancements in AI such as with Sora,  personalised advertising through custom content is set to soar. Existing workflows can be streamlined to enable more collaboration between team members as well as tighter feedback loops. It looks like we can do this cheaply too, so it's going to spark experimentation at new levels across society," he said. 

Agreeing with him, Milind, an AI Scientist from Mercedes, who was expressing independent views, notes that from what has been shared so far, the capabilities of the model seem "quite amazing".

"The consistency and quality of the videos over extended period of time is quite a breakthrough. It would be safe to say that for use cases such as hyper-personalised video content creation it would quite useful. I'm also sure that it will continue to improve offering sound generation and fine-grained control in future," he said.  

Exercising caution around the technology

Saying that, one should not get too excited about the technology too quickly. According to Edwin Yeo, general manager at Strategic Public Relations Group, marketers need to be "wary" of speedily adopting Sora, or they do so with a "big degree of risk". He said:

If there's one thing we learned from advancement in technology, it's that technology tends to outpace regulations and safety concerns.

He added that with Sora and generative AI in general, questions over usage and copyright still remain a big challenge for marketers and content producers.

Apart from copyright and safety concerns, there's also a question still of quality, added Yeo. "It's not great with hands, just like AI art, and there are still questions of the computing power needed to output videos in 4K or 8K formats."

He added that personally, he has been using the likes of Midjourney for concept presentations. Once approved through, he will still go back to photography and DI. "I reckon for the immediate future, Sora will be similarly useful. That's already a big impact in the marketing workflow, but we're very far from Sora being able to replace video production," said Yeo. 

Adding to his point, Rai noted that there are also still a number of potential brand safety concerns that marketers need to be wary of.

"For one, deepfakes and misinformation constitute a new level of risk not seen before which could impact brand safety," said Rai. Additionally, brands may face issues with inappropriate content generation that does not align with brand value or that may be offensive or insensitive. Rai said:

Brand authenticity may take a hit if the world starts to rely on AI generated content and less on human oversight.

Aside from these issues, marketers should also be wary of a lack of human input as AI models such as Sora could misinterpret creative briefs and also create data privacy and security challenges, which may lead to copyright infringement cases.

"Marketers need to use Sora for generating content that resonates with individual preferences and behaviours while placing humans in the part of the loop where the combination of creativity, strategy, analytics and inimitable personal touch shines," said Rai.

Join us this coming 24 - 25 April for #Content360, a two-day extravaganza centered around four core thematic pillars: Explore with AI; Insight-powered strategies; Content as an experience; and Embrace the future. Immerse yourself in learning to curate content with creativity, critical thinking, and confidence with us at Content360!

Related articles:
OpenAI unveils platform for users to make custom GPTs
SG creates new Gen-AI frameworks, asks global input to build trust
What are Singaporeans making of the questioning of TikTok CEO?