Video AI still hasn't had its own "GPT moment" yet.

资讯 / AI 快讯
1

Just saw a 36Kr discussion, and the title was pretty blunt: “From Sora to Kling, video AI hasn’t reached its GPT moment yet.” As someone in the editing game, I’ve already felt it. When ChatGPT dropped for text, it literally changed how we work overnight—you type a few lines and it’s good to go. Video is different.

The tools are getting way more powerful, but we’re still far from “generate whatever you want, stable and controllable, plug-and-play.” Right now, it’s mostly used for B-roll, transitions, or concept clips. If you actually try to use a generated shot in a serious video, you’ll still run into issues like continuity, character consistency, and messed-up fingers or mouths.

I’m not into shouting “this is a game-changer” every time. The tools are still climbing; they help, but they’re not replacing anyone yet. The real moment will come when you can just say one sentence and get a finished clip that’s ready for approval.

2

Lurking.

3

Yeah, honestly the clips you generate now are okay as teasers, but they can’t hold up a whole video. Consistency is way too hard to nail.

4

“GPT moment” is so overused at this point. Every time something new drops, people are like “this is its GPT moment.”

5

Character consistency and continuous motion are the biggest hurdles—stunning in stills, but falls apart the moment things start moving.

6

+1

7

That line hits the nail on the head—stills look amazing but fall apart the moment things start moving. Faces and hands go all wobbly, and continuous motion is the biggest wall to hit.

8

+1 on the character consistency wall — the still frames look amazing, but as soon as they move, it all falls apart.

9

Using snippets as a hook is fine +1, I only dare to use like 2-3 seconds of empty shots when editing too :sweat_smile: