The battle for AI video supremacy just turned into an all-out war with five straight days of releases from one company.
If you thought keeping up with AI was hard before, this week was absolutely exhausting in the best way possible. From native audio in video generation to agents that build workflows for you, everything is accelerating. I just watched a fantastic breakdown by an AI expert who tested every single one of these new tools so we don’t have to guess how they work.
🎥 The Week of Video Overload
The biggest splash came from a Chinese model maker called Kling, which dropped a new AI feature every single day for nearly a week. The standout promises were a specialized editing model, where you can supposedly change textures or motion in existing videos, and a new model featuring native audio generation. Instead of adding sound later, the video creates the audio while it generates the visuals.
🚧 Real-world testing shows mixed results
The industry pro put Kling’s new “01” model through the wringer, trying to turn a car into pizza and replace a book with a glowing tablet to test its editing capabilities. While the tablet swap looked decent initially, the physics fell apart when the object moved, and the “pizza car” didn’t melt as prompted. It seems the promise of editing specific elements inside a video is incredibly exciting, but the technology isn’t quite at the “magic wand” stage yet.
🔊 Sound is coming, but it’s glitchy
Testing the new native audio model revealed some uncanny valley moments; a rapping cyberpunk character had decent lip sync but started repeating words, and sound effects for a Rube Goldberg machine failed to match the visuals. Meanwhile, Runway teased their Gen 4.5 model, which looks visually superior to everything else on the market, though no one actually has access to verify those claims just yet.
🕵️ Agents and secret code names
Beyond video, Google introduced a tool that acts like a smart version of automation software, allowing you to build workflow agents directly inside the Workspace ecosystem simply by describing what you want them to do. On the rumor mill front, OpenAI has a new model internally codenamed “Garlic” that is reportedly outperforming current top models in reasoning, hinting that a significant GPT update could arrive early next year!
There were so many more tests and rapid-fire news updates in this breakdown, including Apple’s surprising entry into video generation. You really need to watch the full video linked below to see these models in action.