
Descript's June Changelog: 70 Features in 48 Hours
Descript shipped its June update following a two-day internal hackathon that resulted in 70 bug fixes and feature requests deployed in 48 hours. The company livestreamed the event, which it called the "Customer-Obsessed Telethon," and donated $14,000 to the Sierra Club based on features completed. According to the official changelog, many of the updates are rolling out throughout the summer as the team completes design and testing.
The update introduces inline tone direction for ElevenLabs V3 speakers, adds Claude Opus 4.8 to the Underlord AI assistant, and redesigns the effects menu with 11 new filters and stackable effects. Smaller improvements from the telethon include transitions that remember last-used settings, an option to restore the classic color theme, an always-on credits and media minutes meter, and support for bullet points in text clips.
Tone Tags for ElevenLabs V3 Speakers
Descript now supports inline tone directions for scripts using ElevenLabs V3 speakers. Tone tags appear as chips directly in the transcript at the insertion point and instruct the model how to deliver surrounding speech, such as whispers, laughs, pauses, excitement, or custom prompts described in plain language.
To add a tone tag, users place the cursor in the transcript and click "Add tone" in the toolbar. A preset picker appears with common options including happy, sad, whispers, laughs, pause, and curious. Custom directions are also supported; ElevenLabs treats them as prompting instructions, and shorter, more specific descriptions tend to produce better results than longer descriptions.
When a tone tag is added or edited, the surrounding paragraph regenerates automatically with the new tone applied. Paragraph breaks reset tone, and the model determines how far the effect carries from each insertion point. Tone tags are exclusive to ElevenLabs v3 speakers. On other speakers, inline notes remain regular annotations.
Underlord Gains Persistent Chat and Claude Opus 4.8
Descript's Underlord AI assistant now maintains persistent chat history per project, allowing users to reference previous requests and build on earlier editing sessions. The update also introduces an automatic review step that reruns edits that miss the mark, reducing the need for manual correction.
As previously reported, Underlord now includes Anthropic's Claude Opus 4.8 model in the model picker. Opus 4.8 is designed for more deliberate, multi-step edits and joins existing models in the assistant's toolset. Users can select the model based on the complexity of the editing task.
Redesigned Effects Menu with 11 New Filters
The effects menu has been redesigned and relocated to the Properties panel. Users can hover over any effect to preview it on the current frame before applying. The update adds 11 new effects to the lineup, including nine stylized looks, stackable effects, and gradient fills.
The redesign consolidates effects into a single drawer, streamlining the workflow for applying and stacking multiple filters. Effects can now be layered, allowing users to combine filters for more complex visual treatments.
Telethon Features and Summer Rollout
The telethon produced 70 bug fixes and feature requests, many of which are deploying in stages throughout the summer. Immediate updates include:
- Transitions memory: Transitions now remember the last-used settings, reducing repetitive configuration.
- Classic color theme: An option to restore the original color scheme for users who prefer it.
- Always-on meter: Credits and media minutes now display persistently in the interface.
- Bullet points in text clips: Text clips support bullet formatting.
Descript also launched the Descript MCP connector in the Claude and ChatGPT marketplaces, along with a CLI update.
The full event recap is available on Descript's Canny page, where the company documented the 48-hour build process. Additional features from the telethon are expected to roll out in subsequent updates.
Sources
1 checkedHow we cover tool news: Create With's tool desk drafts these reports with AI from the sources listed above and checks them against those sources before publishing.




