Welcome to the 【AI Daily】column! This is your daily guide to exploring the world of artificial intelligence. Every day, we present you with the hottest AI news, focusing on developers and helping you understand technology trends and innovative AI product applications.
Discover new AI products Learn More: https://top.aibase.com/
1. Kunlun Wanwei Open-Sources SkyReels-V2: An Infinite-Length Movie Generation Model
Kunlun Wanwei's SkyReels team has launched SkyReels-V2, the world's first infinite-length movie generation model based on a diffusion forcing framework. By combining multimodal large language models and reinforcement learning, the model significantly improves video generation quality and efficiency. SkyReels-V2 not only achieves technological breakthroughs but also expands application scenarios, including story generation and image-to-video, showcasing its broad potential in creative content production.
【AiBase Summary:】
🚀 SkyReels-V2 is the world's first infinite-length movie generation model using a diffusion forcing framework, marking a new stage in video generation technology.
🎬 The model excels in motion dynamics, visual quality, and video length coordination, supporting the generation of high-motion quality and high-consistency videos.
📊 In performance evaluations, SkyReels-V2 outperforms comparable models in several key dimensions, demonstrating its excellent instruction following and video consistency.
Details: https://github.com/SkyworkAI/SkyReels-V2
2. iFlytek's StarFire X1 Significantly Upgraded: Aiming to Compete with OpenAI in the AI Field
iFlytek launched its latest AI model, StarFire X1, on April 21st, aiming to compete with OpenAI's o1 and DeepSeek R1. The model excels in various fields, particularly in education, healthcare, and the legal sector. Despite having fewer model parameters, its overall performance is comparable to industry leaders. Furthermore, StarFire X1's unified "fast thinking, slow thinking" model offers users flexible thinking methods, lowering the barrier to entry for businesses using AI.
【AiBase Summary:】
✨ iFlytek's StarFire X1 improves its generalization capabilities through complex scenario data, making it suitable for industries such as education, healthcare, and the legal sector.
💡 Despite having fewer model parameters than similar products, its overall performance rivals industry leaders, showcasing its strong competitiveness.
🔧 A new model customization and optimization toolchain supports various customization options, simplifying the AI application deployment process for businesses.
3. Unitree Robotics Announces the World's First Humanoid Robot Fighting Competition for 2025
Unitree Robotics will host the world's first humanoid robot fighting competition in 2025, showcasing cutting-edge technology and the thrill of robot combat. The participating G1 humanoid robots, after rigorous training, demonstrate exceptional agility and fighting prowess, particularly their impressive ability to quickly resume fighting after being knocked down. This event is not only a technological showcase but will also drive the development of artificial intelligence and robotics, attracting global tech enthusiasts.
【AiBase Summary:】
🤖 From May to June 2025, Unitree Robotics will hold the world's first humanoid robot fighting competition in Hangzhou, offering an unprecedented visual spectacle.
💪 The G1 humanoid robots, after rigorous algorithm training and hardware debugging, demonstrate exceptional agility and powerful punching ability.
📺 The competition will be broadcast nationwide by China Central Television, allowing viewers to witness the peak duel of robot combat.
4. ByteDance's Coze Space Officially Begins Internal Testing
ByteDance's new AI collaborative office platform, "Coze Space," has entered the internal testing phase, aiming to improve the collaborative efficiency between users and AI agents. The platform features innovative functions such as automatic analysis of user needs, task decomposition, and tool invocation, capable of generating comprehensive result reports. Additionally, the platform introduces an expert agent ecosystem, allowing users to select experts from different fields for in-depth analysis to gain more insights.
【AiBase Summary:】
🤖 Coze Space provides comprehensive services, supporting efficient collaboration between users and AI agents, automatically analyzing needs and decomposing tasks.
📊 It introduces an expert agent ecosystem, allowing users to select specialized agents for in-depth analysis and report generation.
🔧 It supports MCP extension integration, initially supporting multiple tools, and will allow users to publish custom MCPs in the future.
5. Google Releases Gemma 3 QAT Model: Easily Run on a Single 3090 GPU
Google recently released a new version of the Gemma 3 series, specifically the Gemma3 27B model optimized with quantization-aware training (QAT). This significantly reduces memory requirements, enabling users to run large models locally on consumer-grade GPUs. QAT technology incorporates quantization operations during training, minimizing performance loss and improving model performance on smaller devices.
【AiBase Summary:】
💡 The QAT-optimized Gemma3 27B model reduces VRAM requirements from 54GB to 14.1GB, allowing users to run it on consumer-grade GPUs.
⚙️ After 5000 steps of QAT training, the model's perplexity decreased by 54%, maintaining efficient operation on smaller devices.
🌐 Several developer tools such as Ollama, LM Studio, and MLX already support the Gemma3 QAT model, enhancing user experience.
6. Intel Open-Sources AI Playground, Enabling Use of Various AI Models with Intel Arc GPUs
Intel announced the open-sourcing of its generative AI software, AI Playground, marking a significant step in promoting the widespread adoption of generative AI technology and community collaboration. AI Playground is a tool optimized for Intel Arc GPUs and integrated graphics, supporting various generative AI models, allowing users to generate AI images locally and ensuring data privacy.
【AiBase Summary:】
🛠️ AI Playground is a powerful AI tool that supports various generative AI models, including image diffusion models and large language models, ensuring local data privacy.
🌍 The open-sourced AI Playground is released under the MIT license, encouraging developers to freely download, customize, and contribute code, lowering the barrier to entry and promoting community collaboration.
🚀 Intel's open-source initiative is considered a significant breakthrough in the generative AI field and is expected to drive the development of more AI solutions based on Intel hardware.
Details: https://github.com/intel/AI-Playground
7. Reachy2 Robot Released: Natural Interaction, $70,000 Price Tag
Hugging Face, through the acquisition of Pollen Robotics, launched the open-source humanoid robot Reachy2, marking a significant milestone in the combination of humanoid robots and generative AI. Reachy2, with its friendly appearance, advanced sensors, and open-source nature, has quickly become a focus of top laboratories worldwide. The robot not only promotes the mainstreaming of robotics but also provides low-cost innovation opportunities for AI and robotics research, showcasing the huge potential of the future humanoid robot market.
【AiBase Summary:】
🤝 Reachy2 is an open-source humanoid robot launched by Hugging Face after acquiring Pollen Robotics, priced at $70,000.
🛠️ The robot is equipped with advanced sensors and VR remote control operation, supporting flexible programming and customization, promoting the democratization of robotics technology.
📈 Market forecasts predict a $1.7 trillion humanoid robot market size by 2050. Reachy2's open-source model provides innovative opportunities for research and education.
8. ByteDance Research Open-Sources ChatTS-14B: Native Understanding and Reasoning Over Time
ByteDance's research team launched ChatTS-14B, a 14-billion-parameter large language model specifically designed for time-series data, aiming to lower the barrier to entry for time-series analysis through a natural language interface. The open-sourcing of this model has garnered widespread attention, marking a significant advancement in the integration of time-series analysis and generative AI. ChatTS-14B not only provides model weights but also includes detailed documentation and code libraries to assist developers in applications across finance, healthcare, and other fields.
【AiBase Summary:】
📊 ChatTS-14B is a 14-billion-parameter language model designed specifically for understanding and reasoning with time-series data.
🌐 The open-sourced ChatTS-14B allows non-professional users to easily handle time-series tasks using natural language, lowering the barrier to entry.
🚀 The release of this model marks a strategic breakthrough for ByteDance in the AI field, promoting the widespread application of time-series analysis.
Details: https://huggingface.co/bytedance-research/ChatTS-14B
9. Figma Drives AI Revolution: Developing an Intelligent App Builder and Website Creation Tool
Figma is actively expanding into the artificial intelligence field, planning to launch an AI application builder and Figma Sites website creation tool. These new tools aim to quickly generate applications and websites using natural language and existing design resources, lowering the barrier to entry, enabling designers without technical backgrounds to easily build functional applications. Figma's innovations not only enhance the intelligence level of design and development but may also redefine industry collaboration models, despite facing competition from platforms like Webflow and Wix.
【AiBase Summary:】
🛠️ Figma launches an AI application builder, supporting various input formats, lowering the development barrier.
🌐 The Figma Sites tool will help users generate usable websites directly from design drafts, expanding the design ecosystem.
🤖 Figma leverages the Claude Sonnet model to improve its intelligence level, potentially reshaping the collaborative model of design and development.
10. Microsoft MarkItDown MCP: Converts Word, Excel, and More to Markdown Format
In the digital age, Microsoft's MarkItDown MCP (Model Context Protocol) brings revolutionary changes to document processing. This tool supports various file formats such as PDF, Word, PowerPoint, and more, efficiently converting them to Markdown format, greatly facilitating text analysis and the application of large language models.
【AiBase Summary:】
📄 **Multi-format Support**: Supports various file formats such as PDF, Word, and PowerPoint, meeting the needs of different scenarios.
🔍 **Intelligent Document Structure Preservation**: During conversion, it intelligently identifies and preserves the core structure of the document, ensuring information integrity.
⚙️ **Plugin Extension Functionality**: Supports third-party plugins, allowing users to expand functionality based on their needs to meet specific document processing requirements.
Details: https://github.com/microsoft/markitdown
