Official Kuaishou tech account, providing real-time updates on the latest Kuaishou technology practices.
CoTyle introduces a novel open‑source framework that generates unique image styles from a numeric style code, eliminating the need for reference images, lengthy prompts, or LoRA modules, and demonstrates superior style consistency compared to existing solutions like Midjourney.
This article explains the over‑optimization problem in GRPO‑based flow models, analyzes why importance‑ratio clipping fails, and introduces GRPO‑Guard with RatioNorm and cross‑step gradient balancing, showing through extensive experiments that it stabilizes training and improves image quality across multiple diffusion backbones and tasks.
The paper identifies imbalanced concept distributions as the main obstacle to arbitrary concept‑combination in text‑to‑image/video generation, proposes the token‑level IMBA Distance and a lightweight IMBA Loss that adaptively re‑weights training tokens, and demonstrates through extensive experiments and a new Inert‑CompBench benchmark that this loss dramatically improves compositional ability without extra data.
KaiFG, Kuaishou's self‑built AI Feature Generator, unifies fragmented feature extraction frameworks, replaces slow C++ compilation cycles with Python‑level development, and achieves near‑C++ performance through Codon‑based compilation, reference‑counted memory management, and aggressive LLVM optimizations, dramatically shortening iteration time.
This article explains the overthinking problem of large language models, introduces the HiPO framework with hybrid data cold‑start and reinforcement‑learning reward mechanisms that let models decide when to think deeply or answer directly, and shows experimental results demonstrating significant efficiency gains and accuracy improvements across multiple benchmarks.
This article details the EMER framework—a Transformer‑based, end‑to‑end multi‑objective ranking system that replaces handcrafted formulas with a learnable AI model, introduces relative‑satisfaction signals and dynamic loss weighting, and demonstrates significant offline and online performance gains in Kuaishou's short‑video recommendation pipeline.
The KAT‑Dev‑72B‑Exp model, an experimental reinforcement‑learning version of KAT‑Coder, achieves a 74.6% performance boost on the SWE‑Bench Verified benchmark, introduces Trie Packing and entropy‑aware advantage scaling, and showcases a decoupled training architecture that dramatically speeds up large‑scale agentic RL training.
This article explains the core challenges of real‑time bidding, traces the evolution from PID to MPC to reinforcement learning, and details how generative reinforcement‑learning techniques such as GAVE and CBD combine diffusion models, value‑guided exploration, and score‑based return‑to‑go to dramatically improve ad‑bid efficiency and revenue.
The MIDAS framework introduced by the Kling Team combines autoregressive video generation with a lightweight diffusion denoising head to deliver real‑time, high‑quality digital‑human synthesis under multimodal control, achieving sub‑500 ms latency, 64× compression, and robust performance across multilingual dialogue, singing, and interactive world modeling tasks.
Kuaishou's Kling-Avatar leverages a multimodal large‑language‑model‑driven two‑stage generation framework to produce minute‑long digital‑human videos that synchronize lip movements, facial expressions, and body gestures with audio, achieving high visual quality, identity consistency, and controllable storytelling across diverse scenarios.
