Kling 3.0 on NeuralBox: audio, 4K, avatars and motion transfer

Kling is Kuaishou's video generator and one of the world leaders in motion quality: people in the frame move naturally, physics holds together, and the camera behaves like a real camera operator's. The current version, Kling 3.0, is already available on NeuralBox — with audio, resolution up to 4K and two signature modes that most competitors don't have.

This post covers what's new in 3.0, how the Avatar and Motion Control modes work and how much it all costs.

What's new in Kling 3.0

The headline features are audio and 4K. Kling 3.0 generates 5- and 10-second clips with a soundtrack (ambient noise and the atmosphere of the scene), and 4K mode gives maximum picture detail. Motion quality is, as usual, the best in its class: 3.0 follows the prompt noticeably more closely than 2.5 and 2.6.

Both classic modes are supported — text-to-video and image-to-video — with prompts up to 2,500 characters. Aspect ratios: 16:9, 9:16 and 1:1.

Avatar — a talking character from a single photo

Kling Avatar turns one portrait and an audio track into a talking video: upload a photo and a speech recording, and you get a character with natural facial expressions saying your text. The voice track can run up to 5 minutes — long enough for a full review, a lesson or an ad, not just a clip.

Pricing starts at 2,000 tokens per second, which is from $0.07/sec on the Basic plan. A one-minute talking avatar costs from about $4.

Motion Control — motion transfer

The second signature mode: upload your own video with the motion (a dance, for example) and a picture of the character, and Kling transfers the motion from the video onto the character. No need to describe the choreography in the prompt: the model simply repeats what it sees in the reference.

Motion Control is billed per second of the uploaded motion video, rounded up to a whole second: 5,050 tokens (≈ $0.17) per second for Kling 3.0 at 720p, and 2,750 tokens (≈ $0.09) per second for the cheaper Kling 2.6 version.

Pricing

There is no subscription: you top up your balance by card (via Stripe) or with crypto and pay per clip. Prices below are for a 5-second video on the Basic plan ($5 = 150,000 tokens). A 10-second clip costs twice as much, and on higher plans tokens are up to 40% cheaper.

Refunds on failure — unlike the official Kling

The official Kling is sold as a subscription from $6.99/month: credits expire at the end of the month, and failed generations are not refunded. NeuralBox works the other way around: if a generation fails, the tokens go back to your balance automatically, so experimenting with a prompt costs you nothing.

A detailed overview of the model, prompt examples and an FAQ are on the Kling on NeuralBox page.