OpenAI has launched Sora, its new video-generation tool that promises to reshape how creators produce videos using artificial intelligence. The tool is available to selected users and is making waves in the AI space, offering both video creation and editing capabilities that go beyond typical generative tools.

YouTuber Marques Brownlee, who gained early access to Sora, shared his thoughts in a detailed review. He noted that Sora is a separate platform from OpenAI's chatbot, ChatGPT, living on Sora.com. The homepage features a scroll of AI-generated videos, curated by OpenAI, which users can save and organize into folders. Videos can be made from text prompts and uploaded images, providing flexibility for creative projects.

One of the standout features of Sora is the "Re-mix" tool, which allows users to edit existing videos. Users can request specific changes to a video, adjusting how drastic the modifications are with a "strength" setting. The system generates high-quality footage up to 1080p, although the higher the resolution, the longer the video takes to generate. At 1080p, the video generation time can be several minutes, with 720p taking half as long and 480p being the fastest option.

However, despite the tool’s potential, Sora is not without its flaws. Brownlee observed that it suffers from issues typical of AI video generation, such as object permanence problems. In Sora videos, objects can move in illogical ways or disappear and reappear unexpectedly. Additionally, human figures, especially those with legs, often suffer from anatomical inconsistencies.

Sora’s creators have included safeguards to prevent misuse. The platform prohibits the generation of explicit content, violence, or videos featuring minors. Videos made from images of public figures or recognizable logos are also restricted. Each generated video carries a watermark to ensure traceability.

Despite these limitations, Sora is an excellent tool for certain creative uses. Brownlee highlighted its strengths in creating animations, abstract visuals, and title slides. However, for photorealistic videos, Sora’s current iteration still leaves room for improvement. "You can quickly tell it's AI-generated," said Brownlee, referring to the unnatural quality that often characterizes AI-generated content.

While Sora’s potential is undeniable, OpenAI is proceeding cautiously with some features. One notable restriction is the ability to generate videos using real photos or videos of real people. OpenAI is rolling this feature out to a select group of users only and has warned against possible misuse, especially in creating deepfake-like content. The company has adopted a conservative approach to learning from early user patterns and is refining its safety protocols.

Sora also includes metadata to trace the origin of generated videos, which can help platforms detect AI-created content. OpenAI has incorporated measures to avoid copyright infringement, including prompt re-writing to prevent Sora from mimicking living artists' styles, following complaints from artists regarding AI tools.

Despite concerns over potential misuse, OpenAI’s Sora opens up new possibilities for content creation. As AI video generation continues to evolve, it’s clear that tools like Sora will become increasingly important for creators in various industries. However, the fine line between innovation and ethical challenges remains a critical issue for the AI community to address.

TECHCRUNCH