Originally demoed in February 2024, OpenAI’s text-to-video generator SORA is set to publicly launch later this year. SORA is able to produce incredibly realistic-looking videos from short text prompts of up to a minute long.
Watching some of the videos showcased by OpenAI confirms that SORA is the most advanced AI video generator we’ve seen to date. The lighting, positioning and camera motion in all of these clips is very impressive, with a fluid and natural feel throughout.
In one video, created using the prompt ‘A litter of golden retriever puppies playing in the snow’, the snow looks fluffy and crisp while the motion of the fur on the puppies themselves rivals most human-assisted CGI efforts. Another, featuring an astronaut in a salt desert against a rich blue sky, renders extremely realistic facial features while capturing the hyper-stylised look of a movie trailer or high concept TV ad.
The announcement of SORA looks set to completely upturn the world of video creation, sending shockwaves through the film industry. TV adverts and trailers that currently require multiple days of shooting and highly skilled technical crew and editing teams, not to mention storyboarding, design, casting and costume departments, could rapidly become a thing of the past if SORA delivers on its initial promise. As well as the potential for lost jobs, there are also major concerns about how it could be used to create deepfakes and spread misinformation.
Some flaws remain, such as inconsistencies between scenes with objects disappearing and clothing movement looking unnatural. As with many generative AI models, SORA also seems to struggle with creating realistic hands.
While there is currently no date set for a wider release of SORA, we will be closely monitoring for new developments. In the meantime, you can see a selection of videos generated by the model below.
ADD WORKGROUP