What Impact Can AI-Generated Videos Bring to Us?
-
2023, as the inaugural year of AI technology, has seen AI evolve from a creative assistant tool to an independent creator, spanning AI-generated art, AI content creation, and now AI-generated videos. In terms of technical principles, AI-generated videos are fundamentally similar to AI-generated art, but the challenge lies in synthesizing multiple frames into a coherent video, with the greatest difficulty being the control of camera angles and content. This article explores what AI-generated videos look like at this stage and their potential impact on us.
The Quality of AI-Generated Videos
This article uses Google's Imagen Video as an example. It can generate 24-frame 720P video sequences. Beyond understanding text and producing works in corresponding styles, it also accurately comprehends the 3D structure of objects, adhering to real-world physics, making AI-generated videos truly usable.
For instance, asking AI to generate a video of "a teddy bear running on the streets of New York" results in camera work that largely aligns with public expectations, and the quality is quite decent.
Challenging it further, generating a video with keywords like "set on an alien planet, an incredibly detailed sci-fi scene, market perspective, pixel art" yields a style reminiscent of Minecraft. The scene and camera work are impressive, though clarity issues slightly detract from the experience—a problem that could be resolved with better assets and computational power.
How Will the Maturity of AI-Generated Video Technology Affect Us?
From the current technological capabilities of leading companies, AI-generated videos are still in an early "high-definition asset provider" phase. While they can produce desired content, limitations in asset quality and computational power mean the results are merely passable.
Once the technology matures, creators could use lengthy scripts to generate desired content with AI, saving significant time in video production and editing. AI might also emerge as an independent creator. Similar to ChatGPT, if AI can learn human camera techniques and editing styles, it could produce high-quality content at high frequencies—a devastating blow to low-quality creators.
For viewers, in the short term, video platforms may be flooded with AI-generated "junk content," leading to aesthetic fatigue—akin to the current oversaturation of text-to-speech videos featuring generic characters like "Xiao Shuai" and "Xiao Mei." However, in the long run, the maturation of AI technology could bring a "content dividend," ensuring audiences consistently enjoy high-quality content.