Article

Explore our curated collection of AI articles covering artificial intelligence, machine learning, AI tools and more

183 Article•Page 1 / 21

StereoPilot: Efficient and High-Fidelity 3D Video Conversion

StereoPilot enables efficient and high-fidelity 3D video conversion by directly synthesizing the target view without relying on explicit depth maps or iterative diffusion sampling.

Open source

FlashPortrait: Infinite length, high-fidelity portrait animation that preserves identity.

FlashPortrait is a portrait animation system that offers unlimited length, high fidelity, and identity preservation. It's an end-to-end video diffusion transformer capable of synthesizing infinitely long videos while maintaining identity, achieving up to a 6x speedup in inference time.

Open source

Qwen-Image-Layered: Qwen's Open-Source Image Layering Model

Qwen-Image-Layered is an open-source layered image model developed by Qwen that decomposes images into multiple RGBA layers.

Open source

LongVie 2: Multimodal Controllable Ultra-Long Video World Model

LongVie 2 is a multimodal video generation model producing 3–5 minute videos using global guidance, dense and sparse controls, degradation-aware training, and history-context modeling for improved ...

Open source

MapAnything: Universal Feedforward Metric 3D Reconstruction

MapAnything is a feed-forward transformer model that directly regresses metric 3D geometry from various inputs, supporting over 12 reconstruction tasks like multi-view stereo, monocular depth estim...

Open source

IC-Effect: Precise and Efficient Video Effect Editing via Contextual Learning

IC-Effect enables precise, efficient video effects editing using in-context learning with an instruction-guided DiT framework, supporting complex effects like fire and particles while ensuring spat...

Open source

TurboDiffusion: A Video Generation Acceleration Framework

TurboDiffusion, a video generation acceleration framework, can speed up end-to-end diffusion generation by 100∼200× on a single RTX 5090 while maintaining video quality.

Open source

HY-World 1.5: Interactive World Modeling with Real-Time Latency and Geometric Consistency

HY-World 1.5 enables interactive world modeling with real-time latency and geometric consistency.