SkyReels V4

Artificial Analysis Global Leaderboard - Champion

SkyReels V4 achieved #1 ranking on the authoritative Artificial Analysis text-to-video (with audio) global leaderboard, surpassing internationally renowned models like OpenAI's Sora 2 and Google's Veo 3.1, firmly establishing itself in the global top tier. Moving from "generating clips" to controllable, continuous complete video production.

SkyReels V4 Latest Upgrade Highlights

From global #2 to global #1 in just one month, SkyReels V4's capabilities have reached new heights. This upgrade is not just minor tweaks, but a comprehensive capability leap.

Full-Modal Reinforcement Learning System Upgrade

The fully upgraded full-modal reinforcement learning system makes videos "make sense." The model no longer mechanically "assembles frames" according to prompts, but begins to understand the logic of the entire process.

• Full-Modal Semantic Reward Model - Provides the model with a "global evaluation standard," not just checking if individual frames are good, but whether the entire video is reasonable
• Progressive Curriculum Reinforcement Learning Path - From three dimensions: resolution & duration, task complexity, and data difficulty, progressively mastering complex capabilities

New Keyframe Reference Capability NEW

Provide multiple keyframes to the AI, with key transitions entirely under your control, while the model automatically generates the intermediate frames. Emphasizes control over narrative rhythm and action continuity.

New Grid Reference Capability NEW

A feature specifically designed for short dramas (including AI comic short dramas). Users can upload up to 9 plot keyframes at once, and the model will stably extract and preserve character features and scene styles, generating narrative videos with complete logic and consistent characters and scenes throughout. Mainly used for locking character consistency and visual style.

SkyReels V4 Core Features

SkyReels V4 adopts a video/audio dual-stream MMDiT architecture, making it the world's first unified audio-video AI model. It achieves tight integration of audio and video from the ground up, supporting cinematic audio-video synchronized generation.

SkyReels V4 Technical Innovation

Symmetric Dual-Stream MMDiT Architecture

SkyReels V4 adopts an innovative video/audio dual-stream MMDiT (Multimodal Diffusion Transformer) architecture, fundamentally welding audio and video together from the ground up. This architecture allows the model to "see, hear, and create" within a unified framework, fundamentally solving the problem of separate audio-video processing in traditional methods.

Full-Modal Reference Unified Framework

Unifies all input forms including text, images, video, audio, masks, and more under a single framework. Previously, tasks requiring multiple models接力 and manual alignment can now be completed in one generation pass, significantly reducing engineering complexity.

Based on Tiangong Language Model

SkyReels V4 is developed based on Kunlun Tech's self-developed Tiangong language model, inheriting powerful semantic understanding capabilities to accurately understand user creative intent and transform it into high-quality video content.

API Service & Published Research

SkyReels V4 is available through the official website platform and API services. Research papers have been published on arXiv, providing valuable learning and research resources for global developers and researchers. Note: V2 and V3 versions are open-source, while V4 is available through the API platform.

SkyReels Version Comparison - V3 vs V4

From SkyReels V3 to V4, Kunlun Tech AI has achieved major technological breakthroughs. Here's a detailed comparison of both versions:

SkyReels Development History

About Kunlun Tech AI

Kunlun Tech Group

Kunlun Tech Group (Stock Code: SZ300418) is a leading Chinese internet platform company with multiple well-known AI products including Tiangong AI. The SkyReels series is developed by the Skywork AI team under Kunlun Tech Group.

Tiangong AI

Tiangong AI is a large language model independently developed by Kunlun Tech, with powerful semantic understanding and generation capabilities. SkyReels V4 is developed based on the Tiangong language model, inheriting its excellent language understanding abilities.

DramaWave - AI Version of Netflix

Tiangong AI applies SkyReels-V4 to its own short drama platform DramaWave. As an overseas paid short drama platform launched in October 2024, DramaWave has exceeded 80 million monthly active users, achieving a complete closed loop from technology to product to commercialization.

Application Scenarios

SkyReels V4 is widely used in:

• Short Drama Production - Rapid generation of high-quality short drama content, AI comic short dramas in one pass
• Advertising & Marketing - Creating eye-catching advertising videos
• Cinematic Video Content Creation - Professional-level video production
• Game Cutscenes - Standardized, reusable process scaling
• Music Videos - Combined with Mureka platform for one-stop music needs

Frequently Asked Questions

Is SkyReels V4 free?

SkyReels V4 is available through the official website platform and API services. You can experience the online version at skyreels.ai or integrate it into your applications through the API platform. SkyReels V2 and V3 are open-source versions available on GitHub.

How does SkyReels V4 compare to Seedance 2.0?

SkyReels V4 and Seedance 2.0 are both top Chinese AI video generation models. SkyReels V4 currently tops the Artificial Analysis global leaderboard at #1, with its unique advantage being audio-video joint generation capability, making it the world's first unified model to achieve this functionality.

How do I use SkyReels V4?

You can visit skyreels.ai to directly use the online version, or integrate it into your applications through the API platform. For open-source versions, you can get SkyReels V2 and V3.

What input formats does SkyReels V4 support?

SkyReels V4 supports text input (text-to-video), image input (image-to-video), keyframe reference, grid reference (up to 9 images), and other multi-modal reference inputs to meet various creative scenario needs.

What are keyframe reference and grid reference?

Keyframe reference allows you to provide multiple keyframe images to control narrative rhythm and action continuity; grid reference is designed for short dramas, uploading up to 9 plot keyframes to lock character consistency and visual style, generating coherent narrative videos.

Start Using SkyReels V4

Experience the world's #1 AI video generation technology today and start your creative journey. From "generation" to "production," the era of video industrialization has arrived.