Video ChatGPT: platform creates recordings with text – 03/21/2023 – Tech

Starting from a sentence, the artificial intelligence (AI) platform Gen 2, from the startup Runaway AI INC, delivered a short video. It appeared to be drone footage of a desert landscape—just as requested.

The AI ​​is launched on the heels of the success of the ChatGPT text generator robot, which reached 100 million users in two months of operation, and the Dall-E image generator robot, also created by Open AI. Now, with Gen 2, technology has overcome the video frontier.

The presentation of the robot, commanded by the CEO of Runaway AI, Cris Valenzuela, was accompanied by the Bloomberg news agency.

Gen 2 is still in the testing stage. Interested parties can sign up on the Runaway AI website on a waiting list to use the platform. System updates are posted to the Discord chat app.

Runway’s artificial intelligence model started as an image filter. With a base video, it was possible to add a second reference image and create new material.

In this new phase, it is possible to both create videos from scratch and make montages from images and words. According to the company, Gen 2 delivers the best results when there is a sequence of images, such as multiple angles of the same character.

Valenzuela said that in text, simpler commands work better with the Gen 2. In competing text-to-image models Dall-E or Stable Diffusion, the more detail, the better the result.

Still in December, the columnist of the Sheet Ronaldo Lemos had shown that it was possible to produce short videos with automation and contracting services by platform and monetize the content results. The platform that allowed the reading of scripts was Synthesia, capable of creating characters, but without the scene montage.

“Now, anyone anywhere can create entire worlds”, says Runaway’s institutional video released on Monday (20).

Experts, however, are concerned about the spread of fake news with these artificial intelligence models.

Article by FGV (Getúlio Vargas Foundation) professor Luca Belli and researcher from the same institution Nina da Hora for the Sheet showed that ChatGPT allowed the creation of fraudulent emails in series, in a scam known as phishing.

Deep Fake models, which use artificial intelligence to superimpose images onto videos, have already been used to create fake pornographic material of celebrities. Singer Anitta has already been defamed with material from this source.

According to Bloomberg, Runway combines AI and human moderation to prevent users from generating Gen 2 videos that include pornography, violent content or copyright infringement. These methods, however, can fail.

Another defect of the videos generated by Gen 2 are slight distortions, present in all the material on the platform. The details of people and animals also leave something to be desired.

At the presentation of the model, the CEO of Runway, Valenzuela said it was early. “The model will improve over time.”