Can Kling 3.0 Omni not recognize the timbre in the subject reference? Any fix?

Been messing around all afternoon—Kling 3.0’s Omni mode just refuses to pick up the voice from my main reference character, no matter how I tweak it, nothing matches. Weird thing is, switching to the first-and-last-frame mode works fine.

Anyone here actually managed to get this to work? For now, I’m just bypassing it and using the clone voice feature instead—at least that gives me some control, so I’m not rolling the dice every time I generate.

If anyone knows whether the Omni issue is about the reference image specs or if the mode just doesn’t support voice extraction at all, please let me know. I’ve been digging around but can’t find any official word on it.

I got stuck here too—Omni kept failing to pick up the timbre, so I ended up going the cloning route.

first/last frame doesn’t have this issue, which means the Omni chain never even read the audio from the reference, gotta go a separate voice clone route