I just ran a few experiments through Kling 2.0 Pro and none of the generations align with the prompt to the degree that you could easily lipsync it, at all. "Pretty well" doesn't cut it for that, and I've been following the aivideo sub since its inception. There are two models right now that can do convincing lipsync that doesn't look like trash or aligns with the prompt "pretty well": omnihuman/dreamina and veo3. That's it. At most you could run a second pass with something like LivePortrait, but even that is a rung below the quality of SOTA.
That said, I don't need to convince you, you go ahead and see what you want to see.
This latest generation will trigger a seismic shift, not "maybe in the future when the models improve", right now.
Good job buddy. You compared your first few prompt attempts on the RNG machine vs the cherrypicked outputs of other people. But also, I genuinely think you're pulling this argument away from what it was. Tell me if you can see a fidelity improvement from what the last few videogen products that came out. I can link pretty much two videos off-rip from that sub regarding this lipsync thing you seem to be honing in on.
The things you'd profess that "will change the world" have been here. It takes maybe an extra one step, but the quality's been comparable. Yet they haven't 6 months ago. Or a month ago. Why's that? Is it perhaps, that people have a habit of overestimating how much use they can get out of these things in their current state like you are?
That said, I don't need to convince you, you go ahead and see what you want to see.
This latest generation will trigger a seismic shift, not "maybe in the future when the models improve", right now.