The Future of Creative Workflows Is Here!
Tencent unveiled an autonomous AI creative platform, CapCut launched its infinite AI design canvas, and AIRI introduced open-source AI VTubers for gaming and streaming.
This week in AI, creative technology is rapidly evolving beyond standalone design apps into intelligent systems capable of collaborating, remembering context, generating visuals, and interacting like real digital personalities. The focus is shifting toward AI platforms that can create, adapt, and operate autonomously across complete creative and entertainment workflows.
Tencent introduced a new AI creative platform that keeps entire visual projects inside one connected workspace where AI agents can edit visuals, refine assets, understand design intent, and collaborate through specialized creative roles including branding, illustration, UI/UX, video, and 3D design.
CapCut launched Design Studio 2.0, an infinite AI-powered canvas that allows creators to remix ideas, manage references, refine layers, edit with AI brushes, and co-create alongside an AI design agent while receiving twenty free generations daily.
Meanwhile, AIRI introduced an open-source platform for building AI VTubers capable of gaming, streaming, voice interaction, memory retention, and real-time audience engagement across browser, desktop, and mobile experiences.
Together, these developments show how AI is becoming a fully interactive creative collaborator capable of designing, entertaining, communicating, and learning alongside users everywhere today.
Tencent’s New AI Creative Platform Changes Everything
A new AI creative platform is redefining how teams build visual projects by keeping everything inside one unified workspace no constant tool-switching, no broken context, and no scattered assets. The system uses advanced AI agents that understand design intent, reason through creative problems, and autonomously use tools to complete tasks like inpainting, local editing, background separation, and visual refinements. It also introduces a modular skills ecosystem where creators can use prebuilt workflows, build their own custom skills, and share them with the community. What makes it even more powerful is its memory and contextual awareness the agents can read the canvas, learn user preferences over time, and collaborate through specialized expert roles focused on branding, storyboarding, illustration, UI/UX, video production, and even 3D creation. It feels less like a traditional design app and more like an AI-powered creative operating system.
This New CapCut Update Makes Designing Feel Unlimited
CapCut has officially introduced Design Studio 2.0, a major upgrade to its web-based creative platform designed to make visual creation more intuitive, fluid, and collaborative. Built around an infinite AI-powered canvas, the new experience allows users to bring references, drafts, assets, iterations, and ideas into one seamless workspace without breaking creative flow. Users can remix visuals, experiment with endless design directions, edit using AI brushes, refine layers and elements, and even co-create live with an AI design agent. The platform focuses on visual-first creativity, helping users design naturally instead of relying on rigid templates. CapCut says the goal is to let creators “think visually and create freely,” making advanced design workflows accessible to everyone. To encourage adoption, users also get 20 free AI generations every day.
Build Your Own AI VTuber With AIRI’s Open-Source Platform
A new open-source project called AIRI is making it possible for anyone to create fully interactive AI VTubers similar to Neuro-sama. Built using modern Web technologies like WebGPU, WebAssembly, and WebAudio, AIRI supports gaming, live streaming, voice interaction, memory systems, Live2D/VRM avatars, and even local AI inference directly on user devices. The platform allows AI companions to autonomously play games like Minecraft and Factorio, interact with audiences in real time, remember conversations, analyze screens with computer vision, and communicate using synthesized voices. Unlike proprietary AI VTuber systems, AIRI is fully customizable, self-hosted, and open source, giving creators complete ownership over their digital companion. The project also supports browser, desktop, and mobile experiences, making AI-powered virtual personalities more accessible than ever for creators, streamers, developers, and researchers.
Hand Picked Video
In this video, we explore how wastewater treatment plants work, breaking down each stage of the purification process from sedimentation and filtration to biological treatment and showing how contaminated water is turned into clean, reusable water. We also look at a working model of a treatment plant, making it easier to visualize the flow of water and understand why efficient wastewater management is essential for environmental protection and sustainable living.
Top AI Products from this week
Pancake - Every other AI product is a tool that makes you more productive. A copilot. An assistant. A coworker. Something you use. Pancake makes your company autonomous. Agents with roles, goals, and a heartbeat working while you sleep. You set direction, approve the irreversible, the rest runs. Prepare yourself to be prompted by Pancake.
SpotsNow - Track who's advertising on every podcast, what they're spending, and where their campaigns are running then buy open inventory directly from publishers, all in one place. The competitive intelligence layer for podcast ads.
Pitch - Pitch is an AI-powered presentation platform that helps teams create, collaborate on, and deliver professional slide decks faster while staying fully on brand.
Parrot Speech-to-text API - Introducing Parrot: Ringg’s speech-to-text model for production-grade voice agents. Capture Hindi-heavy and noisy real-world conversations with low-latency inference, stronger transcript quality, and Hindi validation built for downstream workflows.
Angel Match- Angel Match is a database of 125k+ investors that frees you from tedious process of searching for the right investors for your startup.We’ve compiled investor names, investment focuses, past investments, locations, company names, and social media links in one place so you can quickly find investors who are best fit with your business.
Revolte- Revolte helps engineering teams turn ideas into production-ready software faster and more safely using autonomous AI agents. The platform can plan changes, generate code, run security and quality checks, create PRs, support deployments, monitor runtime behavior, and identify risks early while engineers stay in control of key decisions.
This week in AI
ElevenLabs Launches Music v2 - ElevenLabs launched Music v2, an AI music model that can switch genres mid-track while keeping songs musically consistent. The update also improves editing, multilingual vocals, and commercially safe music generation.
Google Launches Stitch Challenge- Google’s Stitch Challenge is now live, inviting creators to build interactive interfaces using Stitch and compete for $10K in prizes.
Genspark Launches AI Slides 5.0- Genspark released AI Slides 5.0 with reusable Slide Skills, faster AI generation, built-in data analysis, and custom expert workflows for creating professional presentations.
Heretic Removes AI Censorship Automatically - Heretic is a new open-source tool that removes safety alignment from LLMs while preserving model intelligence. It works with models like Qwen, Llama, Gemma, and Mistral without retraining or fine-tuning.
Secure MCP Tunnels - OpenAI’s secure MCP tunnel connects private MCP servers to OpenAI products through an outbound-only link, keeping internal systems off the public internet.
Paper Of the day
New research in robotic dexterous manipulation introduces a tactile representation called Center-of-Pressure (CoP), which preserves rich contact information while still transferring reliably from simulation to the real world. Instead of simplifying touch into crude, low-dimensional signals, CoP provides a physically grounded way to encode dense tactile data and is supported by a new calibration method that estimates individual sensor (taxel) orientations without needing ground-truth force labels. In tests on two challenging, vision-free tasks peg-in-hole insertion and ball balancing reinforcement learning policies trained in simulation using CoP achieved zero-shot transfer to a real multi-fingered hand and outperformed both binary-contact and raw tactile baselines. The analysis further indicates that these CoP-based policies implicitly learn task-relevant physical properties, such as object mass, suggesting a promising path toward more capable, touch-driven robotic manipulation.
Read this whole paper 👉 here