Graphic 3D

marți, 28 octombrie 2025

News : Aphelion | Hide to Survive

News : Inpaint4Drag framework with Google Colab demo.

Inpaint4Drag introduces a novel framework that decomposes drag-based editing into pixel-space bidirectional warping and image inpainting. Our method achieves real-time warping previews (0.01s) and efficient inpainting (0.3s) at 512×512 resolution, significantly improving interaction experience while serving as an adapter for any inpainting model.

See this project.

Test with this colab online demo.

News : Lightricks introduced LTX-2

Lightricks introduced LTX-2, an open-source AI model that creates synchronized 4K video and audio in real time. It supports multi-keyframe conditioning, 3D camera logic, and various inputs like text, image, video, and audio for precise, style-consistent video generation up to 10 seconds long at 50 fps. LTX-2 runs efficiently on consumer GPUs with multiple performance modes for different creative needs

News : 13,000x faster even the fastest classical supercomputers.

Today, we’re announcing research that shows — for the first time in history — that a quantum computer can successfully run a verifiable algorithm on hardware, surpassing even the fastest classical supercomputers (13,000x faster). It can compute the structure of a molecule, and paves a path towards real-world applications.

... more on the official google blogger.

News : UltraGen: High-Resolution Video Generation with Hierarchical Attention

UltraGen is a new AI that makes high-resolution videos, even 4K, more efficiently by using a smart attention system that balances local detail and global consistency. This can scale up video quality better than previous AI models or two-step super-resolution methods ...

Read more on the official project webpage.

News : DyPE: Dynamic Position Extrapolation for Ultra High Resolution Diffusion

DyPE (Dynamic Position Extrapolation) enables pre-trained diffusion transformers to generate ultra-high-resolution images far beyond their training scale. It dynamically adjusts positional encodings during denoising to match evolving frequency content—achieving faithful 4K × 4K results without retraining or extra sampling cost.

See the official project webpage.