爱窝啦 AI 日报
每日 AI 行业动态、产品更新、前沿研究、开源项目与社媒信号
Today’s Summary
WeChat beta-tests AI scan-to-shop and visual search, Zhihui Jun slashes bipedal robot pricing below 20,000 yuan, and Doubao watermarks appear directly in game art assets.
Open-source tools are encapsulating professional skills into ready-to-use modules—PPT layouts, video editing, and AI-generated image layer editing now have unified entry points.
Today we'll start with WeChat AI and robot pricing, then explore Video Use and Canva Magic Layers workflows.🔥 Today’s Top 10
1. WeChat Beta-Tests AI Scan-to-Shop and Visual Search
WeChat has opened up a batch of its proprietary AI capabilities. User screenshots circulating on social platforms show WeChat beta-testing AI scan-to-shop and visual search . Features cover visual recognition, physical item search, and content generation 🛒. Currently limited to small-scale invitation testing with no public timeline. WeChat’s self-trained large model appears to be moving toward everyday tool functionality within the client.
2. Zhihui Jun Drops Bipedal Humanoid Robot Price Below 20,000 Yuan
Bipedal robots now priced at 19,999 yuan. According to 36Kr, at the Qiyuan Robotics launch event, Peng Zhihui unveiled the Q1 and T1 personal robots , starting at 19,999 yuan, capable of walking and conversing, shipping October 1st 🤖. Previously, the comparable Unitree R1 sold for 39,900 yuan. Backing Qiyuan is Swancor New Material, whose STAR Market stock has surged over 1,800% in 2025, reaching a market cap of 55 billion yuan. Those interested in testing physical units should watch for first-batch user feedback.
3. Game Studio Ships Doubao AI-Generated Images Complete with Watermarks
Game graphics went live with watermarks still attached. Players on social platforms discovered a game’s art assets came directly from Doubao generation , preserving the signature [AI Generated] product watermark 🎨. The industry secretly using AI for artwork is no secret, but submitting work without training custom Loras or hooking up APIs, directly using public products while keeping watermarks—that’s rare. Designers taking game art contracts might want to reconsider where the line is.
4. Open-Source Project Packages PPT Pro Layouts and Analysis Models into Skills
PPT layout capabilities have been packaged into open-source Skills. Channel compilation shows an open-source project bundling PPT pro layouts and analysis models into Skills 📊. Radar charts, Gantt charts, SWOT and Porter’s Five Forces and other professional analysis models are all included. Having AI do PPT now means more than just generation—you can go straight to consulting-grade layouts. Product managers and analysts needing rapid business presentations should try this template library.
5. Video Agent Tool Lets AI Directly Execute Editing Tasks
Video editing can now be commanded via natural language. Channel introduction states a video Agent tool enables AI to directly execute editing tasks 🎬. Trimming clips, removing waste footage, adding subtitles, color grading, and rendering can all be delegated to the Agent. Unlike Remotion’s code-generation approach, this treats AI as an executing editor. Self-media and short-video teams with high editing volumes should test a workflow version.
6. Windows Platform Sees Local AI Audio Function Integration Workbench
Seven local audio AI functions integrated into a single tool. Channel mentions a Windows platform local AI audio function integration workbench has emerged 🎵. Eliminates the hassle of managing multiple launchers, ports, and result directories. Creators wanting to run local audio generation and processing tasks should check deployment steps.
7. Codex-Generated Images Can Be Directly Layer-Separated for Continued Editing in Canva
Adjusting AI-generated images no longer requires regeneration. Channel testing shows Codex-generated images can be directly layer-separated for continued editing in Canva ✂️. Text, images, and graphics can be individually selected and modified—repositioning, resizing, swapping assets all directly operable. Previously required line-by-line adjustments with Codex, requiring regeneration if tweaks went wrong. Now layer separation allows direct Canva adjustments, saving poster and cover designers considerable rework time.
8. Independent Developer Live-Streams Nearly 5 Hours of Modding Qwen into Decision Model
Someone live-modified Qwen into a decision system. Channel reports an independent developer live-streamed nearly 5 hours of modding Qwen into a decision model 💻. From understanding input and data structures step-by-step to final architecture testing, completely unedited. Developers wanting to see zero-to-production model modification processes can find details in the replay.
9. Alibaba DAMO Academy Open-Sources Abdominal CT Diagnostic Model RADAR
Senior radiologist-level CT diagnostic capability now open-sourced. According to Gorden Sun, Alibaba DAMO Academy open-sourced RADAR , a general multimodal model for abdominal CT diagnosis capable of assisting in detecting multiple organ lesions 🏥. It autonomously learns from over 400,000 contrast-enhanced CT images and corresponding text reports. Demonstrates near-expert analytical capability in routine physical screening and complex clinical symptom assessment. Developers and medical institutions working with medical imaging can download and test on Hugging Face.
10. Film Shot Language Encyclopedia Site Compiles 400+ Shooting Techniques
Video creators and AI video generators now have a reference library. Channel recommendation shows the site has compiled 400+ film shooting techniques with corresponding sample clips 🎥. Camera movements alone include 86 types, plus camera positions, lighting, composition, color grading, and transitions. Each shot explains what it looks like, why it’s shot that way, suitable scenarios, and how to write the Prompt. Video creators and AI video generation users wanting to strengthen shot language fundamentals should browse through.
⚡ Product & Feature Updates
AI Education Tools Start Creating Immersive Subject Experience Courseware
Middle school subject knowledge can now become AI-powered interactive experiences. Channel mentions creators are turning the Long March route, human anatomy, physics instruments, and spatial math principles into immersive courseware 📚. The Long March route courseware received 56,000 likes. Classroom scenarization teaching tool developers can search Douyin for related case studies to see actual effects.
Screenshots Can Directly Generate Webpage and Mini-Program Landing Screens
A single screenshot can now become usable code directly. Channel testing shows Ling-3.0-flash-VL can high-fidelity replicate webpages and mini-program landing screens from screenshots 💻. Apple Store official website, Mixue Bingcheng mini-program homepage—both can reconstruct main structure and visual hierarchy. Clicking portfolio images triggers model-generated introductions and dynamic interactions. Designers and front-end developers needing rapid personal homepage, portfolio, or event page setup should try this starting approach.
◎ Industry Shifts & Personal Impact
Ben Thompson Analyzes TSMC Risk and AI Business Logic
Tech analyst believes US total AI dominance is dangerous for the world. Baoyu recaps podcast content stating Ben Thompson on Invest Like the Best judges if mastering AI enables military dominance, China’s game-theory optimal move is destroying TSMC 🏭. Having lived in Taiwan long-term, he just moved back to the US, and this episode covers AI competition geopolitics, risk transfer behind compute shortages, through to major players’ positions in the AI era. The segment comparing Dropbox to OpenAI’s consumer business model delivers the most value. Industry practitioners tracking AI landscape should listen to the full episode.
⌘ Top Open-Source Projects
mvschwarz/openrig: Multi-Agent Unified Framework
Claude Code and Codex can now run as a unified system. openrig has released a multi-agent unified runtime framework , gaining 114 Stars today, total Stars 976. Suitable for developers wanting to orchestrate different AI Agents into the same workflow for trial deployment.
◉ Social Media Picks
Claude Account Registration Time Now Directly Queryable
Those wanting to know their Claude account registration time can now query via API. Gorden Sun shares visiting claude.ai/api/organizations reveals account registration time 📅. His account was among the first batch of Claude Waitlist approved users, likely why the account has never been banned. Users concerned about account stability can confirm their registration batch.
😄 AI Fun Facts
Game Studio Forgot to Remove Watermark
A newly launched game was discovered by players with Doubao’s [AI Generated] watermark prominently displayed in the bottom-right corner of promotional images. The industry using AI for artwork has long been an open secret, but shipping directly with the watermark still attached , skipping even Lora training and API integration steps—that’s something. Netizens joked: Is this performance art or true cost-cutting efficiency?