This is Sora, a new artificial intelligence from OpenAI that creates realistic videos from texts.

OpenAI decided to go further and make a breakthrough in terms of the artificial intelligence revolution. That’s how it turned out litter, a new system capable of creating realistic videos using text commands. Although this is an all-access trial version exclusively for the company’s own researchers, it is expected to be commercialized soon at the same level as ChatGPT And DALL-Erevolutionizing the sector.

How does Sora work?

Its methodology is very similar to that of ChatGPT and DALL-E in that it will suffice provide a hint or text command so the AI ​​tool was capable of creating very creative videos. This way, you can understand information in natural language because trained to understand human language.

Sora was training with a large video library. This means that it has a great ability to recognize movements, descriptions and any aspect requested, and can also recreate them on video. This way, the AI ​​will know what aspects you mean when you talk about types of people, clothes, accessories or visuals.

In addition to capture movements and vibrations with high accuracySora can create multiple frames in a single generated video, so the characters, setting, and visual style will all be of 100% unmatched quality.

Like GPT models, Sora uses Transformer architecturethereby providing superior scaling performance.

Your shortcomings

As OpenAI notes on its website, Sora may have difficulty accurately simulating the physical landscape of a complex scene, without being able to understand specific cases of cause and effect. So, for example, if a person bites into a cookie, there may not be a bite mark left on the cookie.

Sora also presents technical difficulties, as he can confusion in the spatial details of the message. Thus, you may not be able to distinguish left from right or accurately describe events as they unfold over time. He also has difficulty following the camera.

How can it be used?

Thanks Sora you can specify the character’s movements And create complex scenes, detailing every decorative aspect of both the characters and the landscape. This high precision is suitable not only for people, but also for landscapes, as details such as a wet street with specular reflections can be captured.

Something unique about Sora his level of intelligence and context recognition. Thus, if you are asked for a video of a woman walking around Tokyo, it will automatically give you oriental features, as well as the appropriate age and clothing, unless otherwise specified.

Getting the Most Out of Sora This will depend on the user’s ability to create hints. difficult. The more detailed the description, the more accurate the final image will be.

At the moment, being in an early stage of development, Sora can only create 60-second videos.. Additionally, there are some challenges in recreating physical features with complete accuracy.

Its availability

As mentioned earlier, Sora is currently only available to OpenAI researchers and is in the lab testing phase. However, the company hopes it can quickly join your product catalogalthough you have to take them measures to guarantee complete safety of users from possible cyber attacks.

Safety

It is this factor that OpenAI attaches great importance to. Members “Red Team” test Sora’s model negatively. They also create tools to help detect malicious content, a classifier that indicates when Sora created a video and include C2PA metadata if the model is implemented in an OpenAI product.

Existing security practices also apply to Sora.. Thus, the text classifier will check and reject text input requests that violate usage rules and policies (extreme violence, sexual content, hate images, and image rights and privacy).

WITH robust image classifiers All frames of each video may be carefully reviewed to ensure compliance with usage policies. It will bring together policymakers, educators and artists from around the world, understanding their challenges and identifying positive use cases.

OpenAI recognizes that, despite extensive research and testing, it is impossible to predict all beneficial and harmful forms of its new technology. From there What Training and understanding of practical use are essential and critical to running increasingly secure AI systems over time..