The Descript Telethon begins, but first... releases

The Descript Telethon kicks off today,

May 14 at 10:00 AM PT

— tune in for a hackathon, bug fix-athon, feature request fulfill-athon with live demos, team interviews, a talent show, and much more.

We’ve closed submissions for now, but you can still leave comments and upvote your favorite ideas on our Telethon board.

Underlord now lets you pin specific parts of your project directly to a request. Click the in the Underlord chat panel to open the context picker and attach files, scenes, timestamps, or individual layers before you describe what you want done.

This is most useful in complex projects where Underlord might otherwise have trouble knowing what you're referring to. Instead of describing the scene, just attach it.

Your Underlord conversations now persist. Refreshing the page, navigating away, or switching projects no longer wipes your chat. When you come back to a project, Underlord picks up where you left off.

Past conversations are accessible from the Underlord panel, listed in reverse chronological order and auto-titled based on their contents. You can browse older chats and reuse prompts that worked well — useful for complex projects where you've built up a lot of context with Underlord over multiple sessions.

A few details on how it works:

We've added an automatic review step after every Underlord editing turn. After Underlord makes a change, a second pass checks the diff against what you actually asked for, looking specifically for things like unintended deletions, edits that missed their target, overcorrections, or operations that went out of scope. If something looks wrong, Underlord is re-invoked automatically to fix it.

You don't interact with this directly—it runs in the background. The practical effect is fewer cases where Underlord does something technically correct but contextually wrong.

Descript now imports a significantly wider range of media formats. These were selected based on how frequently they were hitting unsupported file errors, from most to least common:

The main navigation areas of Descript have a new look. Workspaces are cleaner and easier to move around in, the Drive sidebar has been reorganized for faster scanning, and Settings has been restructured so related controls are grouped more logically.

Enterprise drives now have access to Custom Workspaces, so larger teams can organize projects into tailored spaces (e.g., by department or initiative), with clearer navigation and less clutter as your Drive grows.

Scribe v2 is now the default transcription model across all supported languages. It outperformed our other models in testing, resulting in fewer corrections per minute and fewer capitalization fixes.

What changed:

See Automatic transcription and Supported transcription languages for more.

GPT Image 2 is now available as a model option in Descript's Generate media panel. It's notable for character consistency and performs especially well on text-heavy images and infographic-style compositions.

You can now add bullet point formatting to text clips in the Canvas.

Select a text clip of any kind, and enter a dash [-] and hit Return to format it into a bullet point or bulleted list.

Since the API entered open beta, the following capabilities have been added:

We're constantly improving and adding functionality based on your feedback, so keep it coming!

When your microphone switches automatically mid-session (e.g. because a device reconnects), you'll now see a toast notification confirming what happened. (Rooms overview)

Attempting to join a Room on a mobile device now shows a clear explanation instead of a broken or degraded experience.

The label in the Underlord panel has been updated. The page at descript.com/templates is unchanged.

Since we're dedicating the next two days to bug fixes and feature requests, we'll post a bigger bug fixes update after the Telethon.