Having watched quite a few sample clips of Seedance 2.0, here are a few brief thoughts.
Many people are missing the point. Discussions are mostly about impressive consistency or strong cue word adherence, such as providing very specific cue words: what effect to produce in 0-3 seconds, what effect to produce in 3-5 seconds.
Frankly, this kind of thinking clearly comes from someone outside the film industry, more like someone just playing with AI.
Because both consistency and cue word adherence are essentially things that Veo 3 already mastered six months ago; they're not new capabilities.
The truly terrifying aspect of Seedance 2.0 (if the demo is as impressive as it appears) is this:
It appears to be a video generation tool, but it actually performs the work of a director and editor.
In today's workflow, creating a complete film is already highly streamlined: Storyboard cue words → Storyboard images → Image-to-video conversion.
With the support of Banana Pro's tools, consistency can be achieved just as well as Seedance 2.0, and even more so because each storyboard can be adjusted individually, the controllability is even greater.
The real difference lies not in consistency, but in the ability to handle storyboards and edit.
The same footage, edited by different people, yields drastically different results.
This has little to do with the footage itself; the essence is the difference in editing mindset: how to find editing points between storyboards—even a few frames off can completely change the viewing experience. Anyone with years of editing experience knows what I'm talking about.
This experience is the biggest barrier between film professionals and AI video players, their long-standing competitive advantage.
The truly frightening thing about Seedance 2.0 is that it's directly dismantling this competitive advantage.
The impressive thing about the Seedance 2.0 finished product I've seen isn't consistency, but the incredibly natural, even professional-level, storyboard transitions.
Even more crucially, achieving this effect doesn't require precisely specifying the seconds between cuts; a general, overarching cue is enough.
In the past, achieving the same viewing experience required manually adjusting storyboards, and abrupt cuts necessitated additional footage for smooth transitions. Now, with Seedance 2.0, this step is completely consumed by the model.
I don't know how much content ByteDance fed Seedance 2.0 to push the model to this level.
But at least from this demonstration, ByteDance is definitely on the right track.
#Seedance
