AI decodes DNA, 3D model animator, Hunyuan-GameCraft, Flux Kontext dev, VeriGen, Gemini CLI
Welcome to the AI Search newsletter. Here are the top highlights in AI this week.
AnimateAnyMesh is a new AI framework that can create high-quality animations for 3D models of any shape or size in just a few seconds. It uses a special technique to compress and reconstruct 3D models, allowing it to generate animations that are both realistic and diverse. This technology has the potential to make creating 4D content more accessible and practical. Read more
Tencent’s Hunyuan-A13B is a new open-source AI model that uses only 13 billion active parameters out of 80 billion total, thanks to a Mixture-of-Experts (MoE) design. It matches the performance of much larger models on key benchmarks, and offers both fast and deep reasoning modes while being resource-efficient. Read more
Google launched Doppl, an app that lets you see how different outfits would look on you using AI. You upload a full-body photo, pick outfit images, and Doppl creates a virtual you wearing those clothes—even generating videos to show how they’d look in real life. Read more
Do you prefer to watch instead of read? Check out this video covering all the highlights in AI this week
Google’s Gemini CLI is an open-source tool that brings Gemini 2.5 Pro’s AI abilities straight to your computer’s terminal. It lets developers use natural language to code, debug, automate, and research directly from the command line, with generous free access and full transparency. Read more
Blackforest Labs released FLUX.1 Kontext [dev], a powerful image editing AI model that runs on regular consumer chips. The dev variant is open weights, meaning anyone can use it locally for high-quality image editing. This model offers professional-level editing performance without needing expensive hardware. Read more
Hailuo 02 is a new SOTA video model that excels at prompt understanding, physics, camera control, and coherence. It's based on a new architecture called Noise-aware Compute Redistribution (NCR), which boosts training and inference efficiency by 2.5x, allowing 1080p video generation at a competitive cost. With unmatched efficiency and precision, Hailuo 02 is rated among the top video models in the world. Try it for free today!
Matrix-Game by Skywork AI is a huge AI model for creating interactive game worlds with detailed control over characters and camera movements. With 17 billion parameters, it keeps visuals smooth and consistent while letting developers shape game environments exactly how they want. Read more
Hunyuan-GameCraft is a new AI framework that can generate high-quality, interactive game videos with realistic movements and controls. It uses a novel approach to unify keyboard and mouse inputs into a shared camera representation space, allowing for smooth interpolation between different camera and movement operations. Read more
DreamCube is a new AI model that can generate high-quality 3D panoramas from single-view images and text prompts. It uses a technique called Multi-plane Synchronization to adapt 2D image models to work with 3D panoramic representations, allowing for more realistic and diverse generated content. This technology has the potential to improve applications such as virtual reality and 3D scene generation. Read more
Researchers have discovered a way to repurpose video generative models to perform controllable image generation tasks, such as converting sketches to images or colorizing black and white photos. This approach, called Dimension-Reduction Attack (DRA-Ctrl), leverages the strengths of video models, including their ability to capture dynamic and continuous changes, to perform various image generation tasks. Read more
A good photo on your Linkedin or business profile makes a huge difference. You could do a physical photoshoot, which costs you over $200 and hours posing awkwardly at a camera. Or, with AI Portrait, just upload one photo, and get a portfolio of 50 professional photos in minutes. Save time and money - try it today!
Researchers have found that multimodal large language models (LLMs) create object representations in a way that is similar to how the human brain does. These models can organize objects into meaningful categories and exhibit semantic clustering similar to human cognition. This suggests that LLMs may be able to develop human-like object representations from linguistic and multimodal data. Read more
VeriGen is the first AI model specialized for generating Verilog code, a programming language used to design computer chips. VeriGen was trained on a large dataset of Verilog code and achieved a 41.9% rate of functionally correct code, outperforming commercial models. This breakthrough has the potential to democratize hardware design and make it more accessible to non-experts. Read more
LongWriter-Zero is a powerful AI model that can generate ultra-long text, up to 10,000 tokens, using reinforcement learning. This model is capable of producing coherent and well-structured text, making it suitable for tasks such as writing stories, articles, and reports. It outperforms other models in terms of quality and efficiency, making it a valuable tool for writers and researchers. Read more