AI Video Slowdown: OpenAI's Sora vs. Runway and Pika

AI Video & Visuals


Screenshots of two videos with OpenAI logo overlaid

F.T.

OpenAI has been showing off Sora, its artificial intelligence video generation model, to media industry executives in recent weeks, building enthusiasm and allaying concerns about the technology's potential to be disruptive to certain sectors.

The Financial Times wanted to test Sora alongside systems from competing AI video generation companies Runway and Pika.

We asked advertising, animation, and real estate executives to create prompts for generating videos they might use in their work. We then asked them for their views on how such technology might change their jobs in the future.

Since Sora is not yet publicly available, OpenAI adjusted some of the prompts before submitting the resulting clips, which it says resulted in higher quality videos.

For Runway and Pika, initial and fine-tuned prompts were populated using state-of-the-art models from both companies. The results are as follows.

Charlotte Bunyan, Arq Co-Founder, Brand Advertising Consultant

OpenAI's revised version of Bunyan's prompt to create a campaign for a “famous high street supermarket”:

Pike and Runway video based on Bunyan's original prompt:

“Although the representation of Sora's people was consistent, the actual visualization of the fantasy playground was faithfully rendered in terms of illustrating various elements that others could not produce.

“It’s interesting that OpenAI changed “children” to “people,” but I’d love to know why. Is that a safety question? Is it difficult to represent the children because they don't have as much training? They chose “the people” rather than white men with beards and brown hair. This is what Sora actually produced, which raises questions about bias.



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *