← Back to Videos

T3-Video: Native 4K Video Generation from a 1.3B Model

Loading video

T3-Video retrofits the forward logic of Wan2.1-T2V-1.3B with multi-scale weight-sharing window attention, generating 3840x2176 video natively with no upscaler; on 4K-VBench it gains +4.29 VQA while running over 10x faster.

Video Generation Text-to-Video Window Attention Wan2.1 Tencent Youtu Lab

Category: world-model

Author: @HuggingApps

Date: 2026-10-08T00:00:00

Duration: 16.2s

Source: https://x.com/HuggingApps/status/2108103602715922897

Reference: https://arxiv.org/abs/2512.13492

Whip-pan transitions: 11 shots in one 30-second AI video

Whip-pan transitions: 11 shots in one 30-second AI video

@Promptwhat · 475 views · 2026-09-28

Video GenerationPromptingWhip Pan

Kling 4.0 announced for October: 4K/HDR, 15 multimodal references, native 30-second generation

Kling 4.0 announced for October: 4K/HDR, 15 multimodal references, native 30-second generation

@Kling_ai · 547 views · 2026-09-28

Kling AIKling 4.0Video Generation

CWM-1: The Most Controllable Interactive Avatar Model

@LemonSliceAI · 523 views · 2026-09-23

World ModelsVideo GenerationDiffusion models

XGEN Generative World Simulation: JING Egocentric Experience Model Plus DAO Shared-World Engine

XGEN Generative World Simulation: JING Egocentric Experience Model Plus DAO Shared-World Engine

@XGEN_labs · 779 views · 2026-09-17

World ModelsVideo GenerationInteractive Video

LTX Native Resolution: AI edits on 4K/8K footage and long shots on a single GPU

LTX Native Resolution: AI edits on 4K/8K footage and long shots on a single GPU

@ltx_io · 396 views · 2026-09-29

LTXVideo Generation4K

Programmable World Model: Executable State Separated From Rendering

Programmable World Model: Executable State Separated From Rendering

@alayastd · 612 views · 2026-09-09

World ModelsVideo GenerationEmbodied AI

As an Amazon Associate, we earn from qualifying purchases.