Kling 3.0 shipped this week, and I ran a set of my standard test prompts against it and against the previous version side by side rather than trusting the launch showcase reel. The gap between generations is real this time, not just an incremental bump dressed up as a major release, and it shows up in a specific place I wasn't expecting.
Camera control finally feels intentional
Previous Kling versions rewarded a specific, almost stilted prompt structure to get reliable camera movement, and deviating from that formula often produced static or barely-moving footage. Version 3.0 responds to more natural camera language, "the camera drifts slowly to the right as she turns," and actually executes something close to what's described rather than defaulting to a locked-off frame. That's the single biggest practical change for how I'll write prompts going forward.
Motion consistency across longer generations
Subject motion staying coherent across a longer clip has been a weak point for every model in this category, and 3.0 holds together noticeably better across the length of a clip than the previous version did, less of that mid-clip drift where a subject's proportions or clothing subtly shift. Still not flawless on complex multi-subject scenes, but a real, measurable step forward on single-subject shots specifically.
- Camera movement now responds to natural language, not just a rigid formula
- Motion consistency across a clip's full length is noticeably improved
- Multi-subject scenes remain the weak point, not fully solved this generation
What I'm changing in my own workflow
My existing prompt templates, built around the older formula, actually underperform slightly on 3.0 compared to more natural, descriptive language, which means I'm rewriting a chunk of my saved prompt library rather than just swapping the model and reusing old prompts. Worth the effort. First impression, cautiously enthusiastic, with the usual note that a launch week's testing doesn't reveal everything a few months of real client work eventually will.