How Claude Design Works 1. Access You go to claude.ai/design. The interface combines chat + visual canvas side by side, allowing you to talk while the design is created in real time. 2. Content Input You can provide input in several ways: - Text: example — “create a pitch deck about X” - File uploads: DOCX, PPTX, XLSX, images - Website capture: simply enter a URL to extract style and elements - Brand: connect your team’s code or design files, automatically applying colors, fonts, and components 3. Design Generation The model (Opus 4.7) creates a complete first version, such as: - Slides - Wireframes - Prototypes - Marketing assets 4. Refinement You can make adjustments without starting over: - Direct comments on elements (e.g., “make this button bigger”) - Edit text directly on the canvas - Adjust using sliders (spacing, color, layout) - Request changes via chat (e.g., “make it more serious”) 5. Sharing and Export - Internal link for your team (view or edit) - Export to: Canva, PDF, PPTX, HTML - Dev integration: send directly for implementation Typical Workflow Idea → Prompt → First version → Adjustments → Sharing → Export/Implementation Key Differentiator Unlike tools like Figma or Canva:you don’t design — you talk, and your brand identity is applied automatically. https://www.anthropic.com/news/claude-design-anthropic-labs

❯ /release-notes ⎿ Version 2.1.110: · Added /tui command and tui setting — run /tui fullscreen to switch to flicker-free rendering in the same conversation · Added push notification tool — Claude can send mobile push notifications when Remote Control and "Push when Claude decides" config are enabled · Changed Ctrl+O to toggle between normal and verbose transcript only; focus view is now toggled separately with the new /focus command · Added autoScrollEnabled config to disable conversation auto-scroll in fullscreen mode · Added option to show Claude's last response as commented context in the Ctrl+G external editor (enable via /config) · Improved /plugin Installed tab — items needing attention and favorites appear at the top, disabled items are hidden behind a fold, and f favorites the selected item · Improved /doctor to warn when an MCP server is defined in multiple config scopes with different endpoints · --resume/--continue now resurrects unexpired scheduled tasks · /context, /exit, and /reload-plugins now work from Remote Control (mobile/web) clients · Write tool now informs the model when you edit the proposed content in the IDE diff before accepting · Bash tool now enforces the documented maximum timeout instead of accepting arbitrarily large values · SDK/headless sessions now read TRACEPARENT/TRACESTATE from the environment for distributed trace linking

Version 2.1.111 Updates Model Command: /model - Model kept as Opus 4.7 (1M context) (default) Key Highlights - Claude Opus 4.7 xhigh is now availableUse /effort to tune the balance between speed and intelligence. - Auto mode is now available for Max subscribers when using Opus 4.7. - New xhigh effort level for Opus 4.7Positioned between high and max. Available via: /effort Improvements - Calling /effort without arguments now opens an interactive slider. - Navigate using arrow keys. - Confirm with Enter. Theme & Interface - New “Auto (match terminal)” themeAutomatically matches your terminal’s light/dark mode.Available via /theme. - /skills menu now supports sorting by estimated token countPress t to toggle. - Ctrl+U now clears the entire input bufferPreviously: deleted to the start of the lineUse Ctrl+Y to restore. - Ctrl+L now: - Transcript footer now shows shortcuts: - The “+N lines” truncation marker is now a full-width rule for better visibility. New Features & Skills - New skill: /less-permission-promptsScans transcripts for common read-only Bash and MCP tool calls and suggests a prioritized allowlist for .claude/settings.json. - New command: /ultrareviewRuns a comprehensive cloud-based code review using parallel multi-agent analysis. Auto Mode - Auto mode no longer requires --enable-auto-mode. Windows, Linux & macOS - Windows: PowerShell tool is rolling out graduallyToggle via: - Linux/macOS: enable with: Permissions & Prompts - Read-only Bash commands with glob patterns (e.g. ls *.ts) - Commands starting with: These no longer trigger permission prompts. Usability Improvements - Suggests closest matching subcommand on typosExample: - Plan files are now named after your promptExample: - Improvements to /setup-vertex and /setup-bedrock: Logs, Debugging & Stability - Headless --output-format stream-json now includes plugin_errors in the init event when plugins are demoted due to missing dependencies. - New environment variable: - Suppressed noisy messages related to: - Reverted the v2.1.110 retry cap for non-streaming fallbackIt reduced wait times but increased failure rates under API load.

Project Overview I built an incredible course about Seedance 2.0. During my research, I had the idea of creating a system that automatically generates prompts to help people create videos and films—without needing to spend so much time learning prompt engineering. Idea & Exploration I went deep into this idea and realized it could become an excellent SaaS product. However, in my case, I see myself more as an explorer than a seller 😄 What I Built Alongside the course, I also created a workflow application to support the process. - It helps guide users step-by-step - It simplifies video and film creation - It reduces the need for technical prompt knowledge - Current Status - The project is still being refined - It is already available on GitHub - The application is currently in Portuguese (but easy to translate) - Target Audience For people who don’t understand filmmaking, this is a great introduction. It lowers the barrier to entry and makes it easier to start creating with AI. https://github.com/inematds/seedance2

🧠 What is DeerFlow 2.0? DeerFlow 2.0 is an open-source project by ByteDance (the company behind TikTok) that works as a “super AI agent” — meaning an AI that doesn’t just respond, but actually executes complex tasks in a real, practical way. ⚙️ Core Idea Unlike a typical chatbot, it: - Plans tasks - Breaks them into subtasks (with sub-agents) - Executes code - Searches the web - Generates complete outputs (reports, apps, etc.) - 👉 It’s closer to an “operating system for AI” than a simple chat interface. 🚀 What’s new in 2.0? Version 2.0 is a complete rewrite (nothing from the old 1.x codebase). Main improvements: - 🧩 Becomes a full-stack “super agent” - 🧠 Long-term memory - 🔁 Handles long-running tasks (minutes or hours) - 🤖 Multiple agents working together - 🖥️ Real code execution in a sandbox (like a “mini computer”) - 🔌 “Skills” system (plug-and-play capabilities) 🧪 What can it do? Practical examples: - Perform deep research and generate full reports - Build websites or apps automatically - Analyze data and create charts - Watch videos + summarize + research further - Generate content (articles, presentations, podcasts) - 🧠 Summary 👉 DeerFlow 2.0 =AI that doesn’t just think → but actually executes complex tasks from start to finish https://github.com/bytedance/deer-flow