A full brush tool is now available in the image editor — paint masks, make precise selections, and touch up generations without leaving the canvas.
The video editor now supports audio tracks — pull sound files from your drive or whiteboard and mix them directly into your videos.
You no longer need to open a whiteboard first — launch the image or video editor directly and start working on assets straight away.
Graphics are now saved with their layer structure intact, so you can come back, tweak one element, and re-export without rebuilding the file from scratch.
Hitting ESC in the editor now prompts to save any pending changes — no more accidentally losing work on a stray keypress.
When a model fails or is unavailable, DFIRST now suggests a compatible fallback you can switch to with one click — your workflow keeps moving even when a provider hiccups.
OpenAI and Seedream image models now stream results as they generate, so you see progress building up in real time instead of waiting for a final image to appear.
The docker panel now has a corner handle you can drag to resize it to fit your screen and your workflow.
Your favorite models now appear front-and-center on the Feed and Whiteboard too — not just in the model selector — so your go-to tools follow you across the app.
Video generations can now produce multiple outputs per run, just like image generations — pick the best take from a batch instead of re-running one at a time.
Social-media scraper outputs (Instagram, TikTok, LinkedIn, etc.) now render natively inside nodes instead of inside iframes — faster, cleaner, and easier to browse.
Deep React optimizations to the whiteboard and AI Agent — noticeably smoother interaction, lower CPU usage, and fewer frame drops on complex projects.
Undo and redo now use a proper command-pattern stack — every action is reversible instantly and reliably, even on large projects with heavy history.
The side navigation now virtualizes the whiteboard list, so opening a workspace with hundreds of projects is just as fast as opening one with ten.
Large file uploads now use multipart transfers instead of base64 encoding — expect dramatically faster uploads for video files, big assets, and Data Room imports.
Opening the docker now puts your cursor straight in the prompt field, and clicking any node instantly opens its configurator — fewer clicks between idea and generation.
Fixed several bugs across the platform. Most notably: Runway Act-Two generation errors, Public Link crashes, asset styling issues, download options restored for Sonar and Gemini Deep Research, audio recordings in the Data Room, and file-rename extension handling.