Kling 1.6 Pro

حول Kuaishou Technology

تأسس في 2011 · Beijing, China

Kuaishou (Kwai outside China) is a major Chinese short-video platform founded in 2011 by Su Hua and Cheng Yixiao, often described as the principal domestic competitor to ByteDance's Douyin / TikTok. Kuaishou's AI Group (KLING team) launched the Kling video model in June 2024, instantly impressing the international community by producing 2-minute photorealistic clips before Sora was publicly available. The Pro tier (Kling 1.0 Pro, 1.5 Pro, 1.6 Pro) is positioned as Kuaishou's premium video model with higher resolution, better prompt adherence and longer durations than the Standard tier. Kling 1.6 Pro (released late 2024) was widely cited as the strongest publicly accessible Chinese video model at its time of launch, ahead of Hailuo and earlier Wan releases.

العمارة

Diffusion Transformer (DiT) with 3D spatio-temporal VAE

Kling is a closed-source latent video diffusion model that Kuaishou describe as a Diffusion-Transformer (DiT) operating on a 3D spatio-temporal Variational Autoencoder. The 1.6 Pro variant generates clips up to 10 seconds at 1080p / 30 fps natively, with extension chains supporting up to 2 minutes. Text conditioning uses a bilingual large-language-model text encoder; image conditioning supports both first frame and end frame for guided shots. The architecture is reported to use 3D attention with full spatio-temporal mixing rather than factorised space/time attention, yielding stronger motion and identity coherence. Training uses a curated video corpus drawn in part from Kuaishou's vast platform data with synthetic dense captions; post-training reportedly combines reward-model alignment for aesthetics with caption-fidelity supervision and selective reinforcement on human-judged quality.

المعاملات

Undisclosed

الإمكانيات

• 1080p / 30 fps clips up to 10 seconds natively, up to 2 minutes with extension
• First-frame and end-frame image conditioning
• Strong bilingual prompt understanding (Chinese, English)
• Director-style camera-control vocabulary
• Lip-sync, motion-brush and elements features in Kling editor
• Available via klingai.com web app, mobile app and API
• Strong character consistency across extensions
• Highly cinematic look on portrait and product shots
• Best for: cinematic shorts, fashion, product video, music videos.

التدريب والترخيص

Closed corpus including licensed video, web video and Kuaishou platform data with bilingual dense captions; exact dataset size undisclosed.

الترخيص: Proprietary commercial licence via Kling AI / Kuaishou terms; commercial use permitted on paid tiers.

اختبارات الأمان: Filters for NSFW, CSAM, public figures and politically sensitive Chinese topics in line with CAC algorithmic-content rules; stricter on Chinese surfaces.