July 21st, 2026
All generated images, videos, and audio files are now served through a global CDN (CloudFront) at full quality, delivering content faster to users worldwide regardless of location.
Faster image downloads — full-resolution image downloads now served via CDN with proper formatting
Faster video downloads — video downloads delivered via CDN with correct media types
Faster video streaming — videos now stream via adaptive HLS for quicker playback start, with native Safari HLS support
Faster homepage loading — homepage assets and thumbnails served from Cloudflare with optimized resolution and preconnect hints
Smarter image sizing — the website now requests appropriately-sized images per device, reducing bandwidth and load times without visible quality loss
Public skills catalog — a curated library of skills is now available directly on the homepage, with highlighted sections and category browsing
Redesigned command menu — the slash-command skill menu has been redesigned for easier discovery and activation
Skill activation indicator — when a skill is active in your agent chat, you'll see a visual indicator and can click to view details
Skills in spaces — skills can now be used in spaces with an improved selection experience
Improved connector menu — the MCP connector selection menu has been improved directly from the prompt box for a smoother experience
Skill editing — skills can be renamed, edited, and managed with auto-normalized names and accent-colored cards
Seedream 5.0 Pro — new high-quality image generation model
WebP optimization — large images are automatically recompressed to WebP for faster loading and uploads
AVIF support — AVIF image uploads now accepted and converted for provider compatibility
Cloudflare Images CDN — all images served through Cloudflare Images with automatic format negotiation and caching
Faster page loads — lazy-loaded media players, asset dialogs, and deferred network queries on the homepage
Higher-quality video delivery — videos are now delivered through an upgraded Mux Plus pipeline for better visual quality, this should match the rate used on Netflix/Youtube
MSE-based playback — smoother streaming with Media Source Extensions
Native Safari HLS — Safari now uses its native HLS engine for better compatibility
Improved video downloads — cleaner download variant handling with CDN delivery
Speed and stability controls now exposed in the audio asset details
Full source audio sent to avatar inference — avatar models now receive the complete source audio for more accurate lip-sync
Audio-driven duration — when duration is missing, avatar generations automatically default to audio-driven timing
Native audio input — Kling O3 models now accept audio input directly for audio-driven video generation
Unified create panels — video, image, and audio creation panels now share a consistent design with shared primitives, accessible from the right sidebar
Rerun quick actions — quickly re-run any image, video, or audio generation from the canvas contextual toolbar
"Create speech" renamed — the canvas audio widget now says "Create speech" for clarity
Improved drag-and-drop performance — smoother canvas interactions
Collapsible agent widget — minimize the agent chat to a compact widget when you need more screen space
More reliable streaming — the agent now reconnects on mid-stream errors instead of failing
Tool call visuals — improved rendering of tool calls including skill activation indicators
Safari support — agent now works correctly with Safari's idle callbacks
Faster skill loading — public skills served from a database-backed catalog cached per pod for instant access