Welcome to the "AI Daily" column! This is your guide to exploring the world of artificial intelligence every day. Each day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and innovative AI product applications.
Fresh AI products Click to learn more:https://top.aibase.com/
1、ByteDance's EX-4D Shakes Up Open Source: Turn Monocular Video into Free Perspective 4D Movies Instantly
EX-4D is a 4D video generation framework developed by ByteDance's PICO-MR team, capable of generating high-quality, multi-view 4D video sequences from monocular videos. This technology solves the challenges of multi-view generation in traditional video generation techniques through depth-closed mesh (DW-Mesh) and lightweight adaptation architecture, and outperforms existing methods in performance metrics.
【AiBase Summary:】
💡 EX-4D uses depth-closed mesh (DW-Mesh) to achieve high-quality generation from monocular videos to multiple perspectives.
🔍 Using rendering mask and tracking mask strategies to solve the problem of scarce multi-view data.
🚀 It comprehensively surpasses existing open-source methods in FID, FVD, and VBench metrics, demonstrating excellent performance.
Details link: https://github.com/tau-yihouxiang/EX-4D
2、Bilibili Opens Anime Video Generation Model AniSora V3 Edition, One-click Generate Various Styles of Anime Video Shots
Bilibili announced a major update to its open-source anime video generation model AniSora V3, significantly improving generation quality, motion smoothness, and style diversity. This version is based on the CogVideoX-5B and Wan2.1-14B models, combined with reinforcement learning and human feedback (RLHF) framework, supporting various anime styles of video generation, providing creators with more powerful tools.
【AiBase Summary:】
✨ AniSora V3 optimizes with a spatiotemporal mask module, enhancing control capability for animation tasks.
🚀 Supports multi-task processing, including single-frame image to video generation, keyframe interpolation, and lip synchronization functions.
📦 Open-source ecology promotes community collaboration, developers can obtain code and datasets via GitHub.
Details link: https://t.co/I3HPKPvsBV
3、DeepSWE Open Source AI Agent System Rises Strongly, Based on Qwen3-32B
DeepSWE is an open-source AI agent system based on the Qwen3-32B model, trained using reinforcement learning, and achieved outstanding performance in the SWE-Bench-Verified test. The system adopts the rLLM framework and improved GRPO++ algorithm, showing strong learning ability and application potential in software engineering tasks.
【AiBase Summary:】
🧠 DeepSWE is based on the Qwen3-32B model, completely trained through reinforcement learning, and the open-source information has been fully released.
🏆 In the SWE-Bench-Verified test, DeepSWE performed excellently, achieving a Pass@1 accuracy rate of 59%, becoming a top performer among all open-source agents.
💡 It uses the rLLM framework and improved GRPO++ algorithm, DeepSWE demonstrates its strong learning ability and application potential in practical software engineering tasks.
Details link: https://huggingface.co/agentica-org/DeepSWE-Preview
4、ByteDance Opens New Model VINCIE-3B: 300 million parameters, supports context-aware image editing
ByteDance opens the VINCIE-3B model that supports context-aware image editing, which is developed based on the MM-DiT architecture, capable of learning from videos and achieving efficient image editing. Its technical highlights include video-driven training, block-causal diffusion transformer, and triple-agent task training, significantly improving the quality and efficiency of image editing.
【AiBase Summary:】
🎥 Video-driven training: VINCIE-3B uses continuous frames of video to automatically extract text descriptions and image sequences, building multimodal training data.
🧠 Block-causal diffusion transformer: The model uses a block-causal attention mechanism to achieve causal attention between text and image blocks, while using bidirectional attention within the block.
🔄 Triple-agent task training: Training through three tasks, including next-frame prediction, current-frame segmentation prediction, and next-frame segmentation prediction, enhances the model's understanding of dynamic scenes and object relationships.
Details link: https://huggingface.co/ByteDance-Seed/VINCIE-3B
5、Stability AI Open Sources Stable Audio Open Small, Turning Smartphones into Audio Creation Tools
Stability AI, in collaboration with Arm, launched Stable Audio Open Small, a lightweight text-to-audio generation model optimized for mobile devices. This model runs locally on mobile devices, supports offline processing, features high efficiency, low latency, and high-quality output, promoting the transformation of AI audio generation technology towards edge computing and mobile devices.
【AiBase Summary:】
📱 Lightweight design: The parameter count is compressed to 341M, suitable for mobile device operation.
🔊 High-quality audio generation: Supports stereo audio generation, without the need for cloud processing.
🌐 Open-source empowers developers: Follows community licenses, lowers technical barriers, encourages creative applications.
Details link: https://huggingface.co/stabilityai/stable-audio-open-small
6、Google Launches Gemini for Education! Free AI Tools Sweep the Global Education Sector
Google launched a new suite of AI tools called Gemini for Education, based on the latest Gemini2.5Pro model and LearnLM learning large model, offering free, powerful, and efficient learning and teaching support for teachers and students worldwide. The tool covers over 30 functions, supports over 40 languages, aiming to empower educators and students with AI technology to create more personalized and efficient learning experiences.
【AiBase Summary:】
🌍 Global education empowerment: Supports over 40 languages and covers over 230 countries and regions.
📚 Free access: Completely free for all Google Workspace for Education users, promoting educational equity.
🔒 Security and privacy: Strictly follows privacy terms to ensure user data security.
7、Topview Avatar 2 Shocking Release! AI Digital Humans Revolutionize E-commerce Live Streaming, Will the Model Era End?
Topview Avatar 2 brings revolutionary experiences to overseas e-commerce and content creators through breakthrough features and realistic effects. Its innovative AI digital human technology enables natural interaction between products and digital humans, greatly improving video production efficiency and content quality.
【AiBase Summary:】
🌍 Global first AI digital human "wearing" products, achieving more realistic interactive effects.
⚙️ One-click generate customized videos, support multilingual mouth movement synchronization, enhance marketing flexibility.
🚀 Revolutionizing traditional UGC video models, reduce e-commerce shooting barriers, help brands go global.
Details link: https://www.topview.ai/ai-product-avatar
8、Perplexity Launches Max Subscription Plan: $200/month Unlock Infinite AI Productivity
Perplexity launched a premium subscription plan called Max, priced at $200 per month or $2000 per year, aimed at meeting the needs of frequent users and professionals. The plan provides unlimited access to Labs, priority experience of new features, and support for the latest cutting-edge models, marking further deepening in the field of AI productivity tools.
【AiBase Summary:】
🧠 Unlimited Labs queries: Meet the needs of professional users for in-depth research and complex projects.
🚀 Priority access to cutting-edge models: Ensure users are always at the forefront of technology.
🔒 Priority support: Provide dedicated infrastructure and faster customer response times.
9、Cursor Boldly Hires! Core Personnel of Claude Code Join Competitors
Cursor hires two core personnel from Anthropic, marking intensified competition in the AI programming market. Although Anthropic faces talent loss, its business remains strong, with significant growth in revenue and valuation. Anysphere then further enhances product competitiveness with these talents.
【AiBase Summary:】
🧠 Cursor successfully hired core personnel from Anthropic, enhancing technical strength.
💼 Boris Cherny and Cat Wu joined Anysphere, driving product innovation.
📈 Anthropic's business is growing rapidly, with significant increases in revenue and valuation.
10、OpenAI Statement: Robinhood's "OpenAI Token" Has Nothing to Do With Us
The article points out that Robinhood launched tokenized stocks of OpenAI and SpaceX in Europe, but OpenAI clearly stated that these tokens are not its equity and have no cooperation with Robinhood. Although Robinhood provided limited-time offers to attract users, American users cannot participate. This incident triggered a heated market reaction, and Robinhood's stock price once soared.
【AiBase Summary:】
💰 OpenAI emphasized that "OpenAI tokens" are not its equity and have no partnership with Robinhood.
⚠️ Robinhood attracted investors through tokenized stocks, but American users cannot participate.
📈 Robinhood's stock price rose due to this news, reaching a historical high.