First impressions: Sora 2 is full of deepfakes and SpongeBob
Key Takeaways
- OpenAI released Sora 2, a new generative AI video model, exclusively on an invite-only iOS app.
- Sora 2 is considered a true rival to Google's Veo 3 due to its impressive realism in generating videos with dialogue and audio.
- The high quality and realism of the generated content raise significant concerns regarding the potential for deepfakes and misinformation.
- OpenAI implemented strong safeguards, rejecting explicit content and restricting the use of uploaded faces unless through the opt-in Cameos feature.
- The Cameos feature allows users to consent to their likeness being used in generated videos, which is a key, yet unsettling, aspect of the new app.
OpenAI has finally released Sora 2, the next-generation generative AI video model, accessible currently through a free, invite-only iOS application, much to the chagrin of Android users. The author gained access and found the technology both impressive and unsettling, noting its capability to generate realistic videos with dialogue and audio based on natural language prompts, making it the first significant competitor to Google's Veo 3. While much of the current content involves lighthearted memes, the high realism of Sora 2 directly correlates with a significant risk of generating harmful deepfakes and misinformation. OpenAI has implemented robust safety measures, rejecting overtly sexual prompts and banning the use of uploaded images containing faces unless utilizing the new Cameos feature. Cameos requires explicit user consent for their likeness to be used, which is a clever, albeit unnerving, attempt to mitigate deepfake concerns, as demonstrated by the flood of videos featuring OpenAI CEO Sam Altman.
