Seven filmmakers tried OpenAI’s Sora with impressive results

OpenAI punta a Hollywood. It’s no longer a mystery.

Bloomberg wrote it a few days agoreporting the conversations that the company’s CEO, Sam Altmanand COO, Brad Lightcup, would have with directors, actors and “big names” in the film industry.

OpenAI’s intent is to place Sorathe new AI capable of generating videos starting from simple text, in the American dream factory.

Hollywood and Sora, after all, have something in common the same propensity to transform even the craziest and most complex idea into reality. The point is that Sora is much cheaper and faster: it doesn’t need sets, actors, crew or editors.

Artificial intelligence

OpenaAI has unveiled Sora, a new AI for creating realistic videos

by Pier Luigi Pisa

February 16, 2024

AI can do everything by itself. Of course, with some limitations.

For now, for example, we know that:

a) The maximum length of videos that can be generated with Sora it’s one minute.

b) To create a clip of approximately twenty seconds, with a low resolution (720p)it takes several minutes, as OpenAI CTO Mira Murati explained to the Wall Street Journal

c) Sora creates amazing images that though they are not free from defects. In particular, the model seems to have problems with the movements of the limbs: the legs of the walker often intersect in an unnatural way; people and animals suddenly split into two.

d) The model – for now – it is unable to associate audio alle scene generate.

The case

OpenAI, Sora videos aren’t perfect. Why didn’t we realize it right away?

by Pier Luigi Pisa

February 19, 2024

Ma OpenAIthe San Francisco startup that thanks to ChatGpt (and to the investments of Microsoft) has reached a market value of at least 80 billion dollars, prefers to underline the merits of its artificial intelligence that generates videos.

And for this reason asked seven filmmakers to use Sora – which is not yet open to the public – to bring their ideas to life and provide the company with useful feedback to improve the new technology.

The artists involved are:

1. Walter Woodman, Sidney Leeder, Patrick Cederberg – Members of shy kids, a multimedia production company based in Toronto. Walter directed the short film “Air Head”.

2. Paul Trillo – Multidisciplinary artist, writer and director.

3. Nik Kleverov – Creative Director and Co-Founder of Native Foreign, an Emmy-nominated creative agency.

4. August Battle – Musician, researcher, creative activist and multidisciplinary artist.

5. Josephine Miller – Co-Founder and Creative Director of Oraar Studio, specialized in 3D visuals, augmented reality and digital fashion.

6. Don Allen Stevenson III – AR/XR digital artist, speaker and consultant.

7. Alex Reben – Sculptor/Artist and OpenAI Resident Artist.

The collective and kids (made up of the first three artists on the list) produced for example a one minute and twenty second clip called “Air Head” really impressive, in which a yellow balloon acts as a leitmotif to the storyfirst taking the place of people’s heads and then flying over extraordinary events and landscapes, from the F1 grand prix to a group of orcas swimming in the sea.

The images of “Air Head”, if they had been shot by real people, with professional equipment, they would probably have cost a fortune. For shy kids, however, it was enough to provide Sora with the ideas necessary to compose their screenplay.

It seems clear that the shy kids clip – like the other six shared by OpenAI – are the result of professional editing that put together the short videos generated by AI.

“While there are still many improvements to be made to Sora, we are already seeing how this model can help creatives turn ideas into reality” OpenAI wrote on its official blog.

The first “impressions” of the filmmakers involved they were all positive. Even too second Ed Newton-RexCEO of the non-profit Fairly Trained which certifies artificial intelligence models trained on truly public or licensed data.

Newton-Rex called OpenAI’s operation “artistwashing”. “This is what happens – explained the CEO of Fairly Trained – when you solicit positive comments on your generative AI from a group of creators, while you train that AI on people’s work without permission or without paying them”.

In the last months The cases involving OpenAI regarding copyright have multiplied. The company has been accused of training its AI on certain content without asking permission. The New York Times took ChatGpt to the forefront. But Sora could soon end up in the crosshairs.

Recently the Wall Street Journal interviewed Mira Murati, the Chief Technology Officer of OpenAI, i.e. the person responsible for the company’s technological strategy. When reporter Joanna Stern asked where the data Sora was trained on came from, Murati closed like a hedgehog: “We used public and licensed data.” “And therefore also videos from YouTubeyes Facebook o Instagram?” the journalist urged. “I’m not sure,” replied the top manager.

In reality Murati then said to Wsj that OpenAI has an agreement with Shutterstocka large database containing images and videos.

Meanwhile, the artists who have tried Sora say they are enthusiastic. “Sora is great at creating realistic scenes, but what excites us is her ability to realize totally surreal ideas,” said the shy kids collective.

Director Paul Trillowhich with OpenAI technology gave life to a freestyle dancer made of garbage, said that “working with Sora is a liberation. No time limits, budget or permissions: I can experiment with bold and innovative ideas. It is precisely by pushing beyond what has already been seen and realizing the impossible that Sora gives his best.”

OpenAI’s initiative is not just an “advertisement” for a technology, a clever way to accredit Sora among the creative community.

It’s a show of strength.

The San Francisco company is sending a message to the very film industry it is courting. This is what AI is capable of. “To you the choice: with us or against us“.