Yann LeCun: Generative models are not suitable for video processing, AI understanding of video must make predictions in abstract space

Tencent Releases GeometryCrafter: Unlocking Geometric Consistency in Open-World Videos with AI

Tencent has recently unveiled GeometryCrafter, a new AI model released via Hugging Face. This model excels at consistent geometric estimation in open-world videos, quickly gaining attention in the tech community. Leveraging diffusion priors, GeometryCrafter opens up new possibilities for deep understanding and processing of video content, offering benefits to creators and researchers alike.

Meta's New Framework VideoJAM: Enhancing AI Video Models' Motion and Physical Abilities

In the field of video generation, despite significant progress in recent years, existing generative models still struggle to accurately capture complex motion, dynamics, and physical phenomena. This limitation primarily stems from traditional pixel reconstruction objectives, which often favor enhancing visual realism while neglecting motion consistency. To address this issue, Meta's research team has introduced a new framework called VideoJAM, which aims to inject effective learning of joint appearance-motion representations into video generation models.

AI Nutrition Assistant Alma Launched: Redefining Health Tracking with Generative Models

Chinese Research Team Unveils VideoChat-Flash, Boosting Long Video Processing Speed by 100 Times

Traditional video understanding models face numerous challenges when processing long videos, including the complexities of understanding the extended context. Although considerable research has been conducted to enhance video understanding capabilities, effectively overcoming the issues of low training and inference efficiency remains difficult. To address these challenges, the research team utilized HiCo technology to compress redundant parts of video information, significantly reducing computational demands while retaining key information. Specifically, HiCo achieves hierarchical compression of the video by segmenting long videos into shorter clips, thereby reducing processing time.

Google AI Introduces a Fundamental Framework for Scaling Inference Time in Diffusion Models

A research team from New York University, MIT, and Google has recently proposed an innovative framework aimed at addressing the bottleneck of inference time scaling in diffusion models. This groundbreaking research transcends the traditional method of simply increasing the denoising steps, opening up new avenues for improving generative model performance. The framework primarily unfolds from two dimensions: one is leveraging feedback from validators, and the second is implementing algorithms to discover better noise candidates. The research team builds on a pre-trained SiT-XL model with a resolution of 256×256, maintaining 250 fixed denoising steps.

Breaking! AI Scientists Take a New Approach, Using Large Models to Automatically Explore Artificial Life

Recently, scientists at Sakana AI have made groundbreaking advancements in the field of artificial intelligence, successfully utilizing vision-language foundation models (FMs) for the first time to achieve automated searching for simulations of artificial life (ALife). This project is called ASAL (Automated Search for Artificial Life).

Israeli Tech Company Lightricks Launches LTXV! Generate an AI Video in 4 Seconds

Israeli tech company Lightricks has recently launched a lightning-fast AI video processing platform called LTX Video (LTXV). This platform uses advanced artificial intelligence models that can run on commonly used graphics processing units (GPU). Released in late November, LTXV can generate a 5-second AI video clip at a resolution of 768x512 in just 4 seconds, making it stand out among other AI video platforms. With its 2 billion parameters, it is capable of even...

Researchers are Cultivating AI Scientists, and Progress is Better Than Expected

Recently, there has been a new trend in the scientific community — AI scientists. Simply put, this type of artificial intelligence can not only propose hypotheses and conduct experiments but also independently write research papers. While I have always believed that the best role of AI is as a tool for humans, rather than a replacement, this AI scientist seems to show considerable potential. Image source note: The image is generated by AI and is licensed from service provider Midjourney. We often hear about cases of scientists utilizing AI, such as the research conducted by MIT in 2019.

Disrupting Tradition! Lumina-mGPT Can Create Realistic and High-Resolution Images from Text

Multimodal generative models are leading a new trend in artificial intelligence, focusing on the integration of visual and textual data to create multifunctional AI systems that perform various tasks, from image generation to understanding and reasoning across different data types. A key challenge is to enhance the capabilities of autoregressive (AR) models so they can generate high-detail images based on textual descriptions. While diffusion models excel in generating high-quality images, AR models lag behind in image quality, resolution flexibility, and multitasking capabilities. Researchers from Shanghai AI Lab and The Chinese University of Hong Kong have introduced Lum

Sakana AI Launches New Model to Revive Traditional Japanese Ukiyo-e Art

Tokyo-based startup Sakana AI has introduced two new image generation models — Evo-Ukiyoe and Evo-Nishikie. Image from Sakana AI official These models can be found on the Hugging Face platform, with the main purpose of generating Ukiyo-e style art through text or image prompts. As is known, Ukiyo-e is a traditional Japanese art form popular from the 17th to the 19th century, depicting themes such as historical scenes, natural landscapes, and sumo wrestlers. Sakana AI aims to use AI to bring thi