Kling 2.1 lip-sync finally crossed into 'I'd put this in a client video' territory

by @priyawilson_creates in c/kling-community · · Lip-Sync

Did a talking-head corporate explainer, 8 lines of dialogue, 2.1 Master, Pro mode. The lip-sync is the first I've gotten that doesn't scream AI the second someone says a plosive.

Key things that helped: clean front-ish facing keyframe, audio with clear consonants (mumbly VO wrecks it), and keeping each synced clip under 10s. Longer than that and the mouth starts doing its own thing around the edges.

It's not perfect — side angles still get mushy and anything above a calm speaking cadence (laughing, shouting) falls apart. But for straight talking-head? Shippable. That's a sentence I couldn't write six months ago.

309 points · 9 comments

Comments

  • @henryanderson_studio · · 100 points

    'Clear consonants in the audio' is the tip nobody mentions. Garbage-in VO = garbage lip-sync, the model can only sync to what it can hear.

  • @miachen · · 70 points

    I actually re-record VO slightly over-enunciated now specifically for the sync pass, then nobody notices because the lips sell it.

  • @dianapark_art · · 61 points

    Side angles being mushy is the current wall. Anything past 3/4 and the far corner of the mouth smears. For dialogue scenes I'm stuck shooting everyone near-frontal.

  • @charlotteferreirapro956 · · 52 points

    Same. My AI films all look like news broadcasts because every talking shot has to be frontal lol.

  • @dianasmithworks · · 36 points

    Shippable talking-head is a genuine milestone. Congrats, that's real money on the table now.

  • @nikolaitanaka_art · · 18 points

    Does it handle a beard okay? Facial hair used to destroy the mouth tracking.

  • @emmakumar_motion119 · · 29 points

    Short beard fine, big bushy beard still eats the mouth. Mustaches are weirdly cursed.