OpenAI’s Sora and Google’s Gemini aren’t just generating images; they’re being used to test the limits of what’s possible with AI video. The “Will Smith eating spaghetti” test has become a benchmark for generative AI, revealing rapid progress in the field. But as AI models mature, especially those based in the US, stricter guardrails are being implemented to prevent the misuse of copyrighted material and third-party likenesses.
Key Takeaways
- AI video generation has advanced significantly since 2023, with models like Kuaishou’s Kling 3.0 creating increasingly realistic and cinematic videos.
- Major players like OpenAI and Google are implementing stricter guardrails to prevent misuse of likenesses and copyrighted material, limiting the ability to recreate certain tests.
- The “Will Smith eating spaghetti” test, once a staple of AI video generation, is becoming harder to execute due to these restrictions and Hollywood’s crackdown on AI models trained on its intellectual property.
How Far Has AI Video Generation Really Come?
The evolution of AI video generation is remarkable, especially when looking at the “Will Smith eating spaghetti” test. What started as a pixelated mess created with ModelScope has transformed into a cinematic scene, thanks to models like Kling 3.0 from Kuaishou Technology. This Chinese tech company’s AI can generate videos where Will Smith not only eats spaghetti but also engages in conversation, showcasing how much generative video has matured in just three years. While some may still recognize the telltale signs of AI, the progress is undeniable. The advancements are attributed to improvements in neural networks and diffusion models, which allow for more coherent and realistic video generation. According to Statista, the generative AI market is projected to reach $109.8 billion in 2026, indicating continued investment and development in this area.
Why Is It Getting Harder to Recreate the “Spaghetti Test”?
Despite the advancements, recreating the “Will Smith eating spaghetti” test is becoming increasingly difficult, especially with US-based AI generators. Companies like OpenAI, with their Sora model, and Google, with Gemini’s Veo 3.1, have implemented strict guardrails to prevent the use of third-party likenesses and copyrighted material. This is partly due to Hollywood’s crackdown on AI models trained on its intellectual property, as evidenced by the ongoing legal battles between studios and AI developers. Mashable’s attempt to recreate the test using Sora and Veo 3.1 was denied due to copyright concerns, highlighting the limitations now in place. “These guardrails are essential to prevent misuse, but they also stifle creativity and experimentation,” said Dr. Fei-Fei Li, professor of computer science at Stanford, in a recent interview. The tightening restrictions signal a shift towards responsible AI development, but also raise questions about the future of AI-generated content and its potential for parody and satire.
Products/Companies Mentioned
- Sora – OpenAI’s text-to-video model, released in early 2026, known for its ability to create highly realistic and imaginative scenes up to 60 seconds long, currently in limited access.
- Google Gemini – Google’s multimodal AI model, featuring Veo 3.1 for video generation, capable of creating high-resolution videos with cinematic qualities, integrated into various Google products.
- Kuaishou Technology – Chinese technology company, developer of the Kling 3.0 video generator, known for its advancements in AI-powered video creation and social media platforms.
- ModelScope – An open-source text-to-video framework developed by Microsoft, allowing users to create videos from textual descriptions.
What This Means
- For content creators: The increased restrictions on AI video generation mean more focus on original content and less reliance on replicating existing characters or intellectual property.
- For AI developers: Balancing innovation with ethical considerations and copyright compliance is crucial for the long-term viability of AI video generation models.
- For the entertainment industry: The crackdown on AI models trained on copyrighted material could lead to increased collaboration between Hollywood and AI companies to develop mutually beneficial solutions.
