EASYHUB JOURNAL / 2025-05
AI news in May 2025
16 stories, ordered by their original dates. Open a title for its summary, image and source link.
- 新京报
Tongyi Lingma launches a standalone IDE with project context and MCP
Tongyi Lingma’s standalone AI IDE combines its plugin capabilities with an agent that searches projects, runs terminal tasks and calls MCP tools. Inline suggestions and project memory support iterative development. External integrations and commands still operate within the user’s authorized environment.
- Tencent
HunyuanVideo-Avatar opens audio-driven character animation
HunyuanVideo-Avatar opens code and weights for animating character images from audio, emphasizing identity consistency, expressive delivery and broader body movement. It can support creator workflows for talking or singing characters. The output is generated video, distinct from software that tracks a performer live through a camera.
- IT之家
Skywork’s mobile agents combine documents, slides and spreadsheets
Skywork brings research, document, slide and spreadsheet workflows to mobile with office-format exports. It produces editable starting materials, while citations, numerical values and final layout still need user review.
- IT之家
UBTECH and China Telecom present a home desktop AI robot
The collaboration integrates conversation, visual semantics and memory into a desktop robot with smart-home ambitions. Sales targets are plans, and demonstrated capabilities do not establish support for every household service.
- IT之家
Gemma 3n previews a multimodal design for mobile devices
Google previews Gemma 3n’s memory-efficient design for local experiences. Total parameters, effective memory usage and preview availability are separate measurements that should be checked independently.
- IT之家
Tencent Cloud expands its knowledge engine into an agent platform
Tencent upgrades its model knowledge engine into the TCADP agent-development platform, integrating DeepSeek-R1, V3 and web search for enterprise applications. World-scene generation and further enterprise or on-device model releases are announced as future plans, distinct from the platform capabilities available at the event.
- IT之家
Stitch turns prompts and sketches into UI designs and frontend code
Google Labs introduces Stitch as an experiment that turns prompts, screenshots or wireframes into interface designs. Users can refine layouts in conversation, transfer designs to Figma and export frontend code. It helps bridge design and implementation without claiming to deliver a finished backend, database or tested business application.
- IT之家
GitHub’s coding agent turns assigned tasks into reviewable changes
GitHub connects assigned tasks, repository environments and reviewable changes in an agent workflow. Generated pull requests still require review, with logs and tests providing evidence for evaluating the proposed work.
- IT之家
Windows AI Foundry combines local models and inference tooling
Microsoft introduces local model selection and Windows inference tooling across different processors. APIs and hardware integrations have distinct preview stages, so the shared platform name does not establish universal feature availability.
- Google
Google Flow combines video generation with scene and asset controls
Flow combines Veo, Imagen and Gemini in a filmmaking workspace with reusable characters and assets, camera controls and scene extension. It helps connect individual generations into a story. The initial launch is for US Google AI Pro and Ultra subscribers, with Veo 3 audio access depending on the launch plan.
- The Verge
NotebookLM brings source-based reading and audio overviews to mobile
NotebookLM brings source-based questions and Audio Overviews to Android, iOS and iPadOS. Users can import materials and listen in the background or offline. These generated explanations remain grounded in supplied sources rather than independently verifying the underlying information.
- IT之家
Cloud Codex previews coding tasks in isolated environments
Cloud Codex works on repositories in isolated environments and returns changes with execution logs. It is distinct from the local CLI, with the article documenting the preview’s access and networking limitations at launch.
- IT之家
Hunyuan Image 2.0 adds a real-time drawing board and live previews
Hunyuan Image 2.0 opens for registered use on Tencent’s site, with previews responding to prompt changes. A real-time drawing board generates rendered results from sketches and parameter edits, and can combine multiple sketches on one canvas. The announcement concerns online access, not a release of model weights.
- IT之家
Manus opens registration for its task-based agent platform
Manus moves beyond invitation-only access and documents its launch-era task credits and plans. This records a product-access milestone rather than guaranteeing current prices, credits or task outcomes.
- Notion
Notion combines meeting notes, connected search and research in the workspace
Notion AI for Work adds meeting transcription and action items, search across connected tools and a research mode for drafting documents from gathered material. Meeting Notes and Research Mode are available first, while Enterprise Search and connectors roll out gradually, bringing meetings, conversations and project knowledge closer together.
- IT之家
HunyuanCustom generates new video scenes around a reference subject
HunyuanCustom uses an image to establish a subject’s identity while text specifies actions, clothing and surroundings. The initial release opens single-subject generation and an online demo. Multi-subject, audio-driven and local video-editing capabilities are described as extensions, with further open releases planned at the time.