Kling 4.0

• 🎬 Kuaishou's newest Kling generation, here in image-to-video form
• 📏 Single generations run 3 to 30 seconds
• 🆕 Up to 10 keyframes guide beats inside one take
• 🔧 Kling states Omni Reference mixes image, video and depth inputs
• 👁️ Up to 4K output with 10-bit HDR and 21:9 ultrawide
• 💬 Kling states audio is generated synchronized with the video
• ⚡ Kling says motion stability improved for fast action and camera moves
• 🏢 Released October 2026 by Kuaishou, Kling AI's parent

Kuaishou

Kuaishou Technology is a Chinese publicly traded company founded in 2011 by Hua Su and Cheng Yixiao, headquartered in Beijing's Haidian District and listed on the Hong Kong Stock Exchange. Originally known for its massively popular short-video platform —…

Kling 4.0 (image-to-video) is the animation entry point into Kuaishou's fourth-generation Kling video model, which Kling describes as a native multimodal model for video creation with more stable complex motion, stronger reference consistency and more controllable storytelling. It sits alongside Kling 4.0 for prompt-only generation and Kling 4.0 R2V for reference-driven builds; here you supply a starting frame, optionally a final frame and intermediate keyframes, and the model fills in the motion.

The clearest generational change over Kling V3 Pro and the June Kling V3 Turbo Pro refresh is duration and control: a single generation spans 3 to 30 seconds, with up to 10 keyframes placing specific story beats inside that take. Kling also states that 4.0 better preserves subject structure, spatial relationships and motion continuity during fast movement, tracking shots and orbiting camera moves.

On the picture side, Kling reports better highlight and shadow retention in backlit, night and neon scenes, reduced color banding, and added 21:9 ultrawide framing, with 4K and 10-bit HDR output. Kling states that audio is produced synchronized with the generated video.

Compared with earlier tiers such as Kling 2.6 Pro, Kling states the Omni Reference system is broadened to mix image, video, white-model and depth-map inputs for identity and camera guidance.