Kling 2.6 adds custom voice control and complex motion tracking

AI video generators have a dirty secret: characters look great in still frames, but the moment they try to talk, the illusion falls apart. The voices switch tones between clips, the lips desync, and hands often dissolve into a blurry mess.

Kuaishou just rolled out a pair of major updates to Kling 2.6 designed to tackle exactly these headaches: advanced voice control and improved motion tracking.

Custom voices and consistent characters

Kling 2.6 previously introduced synchronized audio generation, aiming squarely at the kind of native sound design Google demoed with Veo 3 and OpenAI teased with Sora 2. The model generates ambient noise, sound effects, speech, and background music in a single pass.

The new Voice Control feature pushes that concept much further. Kling now handles everyday dialogue, narration, singing, and even rapping, supporting both pure text prompts and text-plus-image inputs.

The real breakthrough for creators is custom voice training. Users can upload their own voice samples to train a dedicated profile or directly upload an audio track to drive the video. This finally gives creators consistent character voices across multiple scenes, making serialized episodic content far more practical.

Cleaner hands, sharper movement

Alongside voice capabilities, Kling updated its motion control engine. The model now tracks full-body kinematics with much higher fidelity, holding up even during high-speed actions like martial arts and choreography.

Kuaishou specifically targeted two common failure modes: messy, blurred hands and unnatural lip sync. Creators can upload between 3 and 30 seconds of reference motion video, using text prompts to restyle the scene around the exact movement path.

In terms of pricing, Kling is accessible via its own credit system as well as third-party platforms like Fal.ai, Artlist, and Media.io. API generation runs roughly $0.07 to $0.14 per second of video, undercutting several prominent competitors.

Why it matters

We think Kuaishou made a very deliberate and practical move here. Instead of chasing marginal gains in raw pixel fidelity, the team is eliminating the friction points that make AI video production a nightmare for indie creators.

For solo filmmakers and marketers, the combo of voice cloning and reference-based motion control collapses the production pipeline. Previously, you had to generate silent video clips, run separate text-to-speech models, and manually lip-sync everything in third-party post-production tools. Testing Kling's voice binding is well worth the time if you are building serialized stories or virtual avatars.

Strategically, this highlights Kuaishou's massive competitive moat. As the operator of Kwai, Kuaishou sits on an endless stream of real-world short videos paired with native speech, background soundscapes, and expressive physical gestures. That kind of synchronized multimodal training data gives Kling an inherent edge over pure-play AI labs.

The open question is content quality. As friction evaporates and API costs drop below $0.15 a second, platforms risk being inundated with low-effort, formulaic video spam before creators figure out how to tell genuinely compelling stories with the tech.