Pinned Tweet
You thought of a better intro after you stopped recording. Everyone does.
Type it in and Descript reads it in your voice.
That voice now runs on Eleven v4, our new default AI speech model, so your accent and pacing come along too.
Type it in and Descript reads it in your voice.
That voice now runs on Eleven v4, our new default AI speech model, so your accent and pacing come along too.
Introducing Eleven v4 and Eleven v4 Turbo, our fastest and most emotive voice models yet.
Ranked #1 by Artificial Analysis.
14
0
23
2,749
How HubSpot's Starter Story team repurposes one video into dozens x.com/i/broadcasts/1…
1
0
4
1,162
That video you keep meaning to make? Give it a morning.
Make Content, Not Excuses returns to Descript’s SF HQ on Oct 6th for #SFTechWeek. Write, record, edit, ship. With help from our team. Leave with a video ready to post.
Join us: partiful.com/e/2DaT8iADDtBm…
Make Content, Not Excuses returns to Descript’s SF HQ on Oct 6th for #SFTechWeek. Write, record, edit, ship. With help from our team. Leave with a video ready to post.
Join us: partiful.com/e/2DaT8iADDtBm…
The official SF + LA Tech Week Calendar is LIVE!
Over 2,300 events, most free to attend. Apply now before they fill up!
tech-week.com/calendar
0
1
5
1,486
You have one move in your process you've never explained to anyone.
This week, teach it. Descript + @contra are running a challenge: teach one thing from your workflow, made in Descript. $5,000 for the best one.
Enter here: bit.ly/46xwDg9
This week, teach it. Descript + @contra are running a challenge: teach one thing from your workflow, made in Descript. $5,000 for the best one.
Enter here: bit.ly/46xwDg9
1
1
4
1,577
Most tools try to do the editor's entire job. HubSpot's Starter Story team built one that only does the repetitive part.
Hub & Spoke, on Descript's API: one video → 55 clips → 1.35M+ views.
Sept 9, we get into how.
Register → bit.ly/3Ui5LOB
Hub & Spoke, on Descript's API: one video → 55 clips → 1.35M+ views.
Sept 9, we get into how.
Register → bit.ly/3Ui5LOB
1
0
5
2,036
New! Captions that earn their screen time.
18 new caption designs, now in Descript—sorted by theme, from clean and quiet to loud and fun.
See them in action here, then add them to your video: bit.ly/45TWOxq
18 new caption designs, now in Descript—sorted by theme, from clean and quiet to loud and fun.
See them in action here, then add them to your video: bit.ly/45TWOxq
1
0
4
1,335
You could stop doing all five of these by hand by Friday (that's tomorrow):
→ sales call → sizzle reel
→ webinar → filler-word-free cut
→ one video → ten languages
→ blog post → video
→ raw footage → social clips
Ask for it in Claude or ChatGPT. Descript does the rest, ign up at bit.ly/4gczctO
→ sales call → sizzle reel
→ webinar → filler-word-free cut
→ one video → ten languages
→ blog post → video
→ raw footage → social clips
Ask for it in Claude or ChatGPT. Descript does the rest, ign up at bit.ly/4gczctO
1
1
3
1,223
Most AI labs build general-purpose generative models. We don't.
We take hard, specific problems in the editing workflow and build models that solve them. We're hiring researchers to do more of it.
Some of what we've worked on lately:
- Fix a wrong word without re-recording it. We trained our own continuous neural audio codec and a tightly coupled flow-matching generator for zero-shot speech regeneration.
- Audio-driven lip sync. When you edit audio, the speaker's lips stop matching the words. We trained a codec + flow-matching model that regenerates only the facial regions that need it, keeping identity, lighting, and temporal consistency intact.
- Jump cut smoothing. Cut part of a video and you usually get an obvious visual jump. Our model synthesizes a physically plausible bridge between the remaining clips, so the edit reads as one continuous take.
That's some of the recent work.
What's next is the bigger swing: multimodal understanding and agentic video editing—models that watch, listen, reason about, and eventually edit long-form media on their own. That's the work we're hiring for.
The team's research: descript.com/research
Apply: job-boards.greenhouse.io/descript/jobs/…
We take hard, specific problems in the editing workflow and build models that solve them. We're hiring researchers to do more of it.
Some of what we've worked on lately:
- Fix a wrong word without re-recording it. We trained our own continuous neural audio codec and a tightly coupled flow-matching generator for zero-shot speech regeneration.
- Audio-driven lip sync. When you edit audio, the speaker's lips stop matching the words. We trained a codec + flow-matching model that regenerates only the facial regions that need it, keeping identity, lighting, and temporal consistency intact.
- Jump cut smoothing. Cut part of a video and you usually get an obvious visual jump. Our model synthesizes a physically plausible bridge between the remaining clips, so the edit reads as one continuous take.
That's some of the recent work.
What's next is the bigger swing: multimodal understanding and agentic video editing—models that watch, listen, reason about, and eventually edit long-form media on their own. That's the work we're hiring for.
The team's research: descript.com/research
Apply: job-boards.greenhouse.io/descript/jobs/…
2
2
7
1,269
You don't micromanage your video edits anymore. You just ask.
Underlord does the heavy editing and sweats the details—like catching that B-roll was landscape in a portrait video, and reshaping it before you notice.
Reach it from your assistant (Claude, ChatGPT) or build it into imeline edits required.
Sign up at bit.ly/4gczctO
Underlord does the heavy editing and sweats the details—like catching that B-roll was landscape in a portrait video, and reshaping it before you notice.
Reach it from your assistant (Claude, ChatGPT) or build it into imeline edits required.
Sign up at bit.ly/4gczctO
2
0
2
1,250
Opus 5 shipped this morning. It's in Descript same day.
Underlord doesn't wait for Monday.
Underlord doesn't wait for Monday.
We've been testing Opus 5 at @descript for a few days and have been really impressed. It's better *and* more efficient than 4.8, putting @claudeai Opus back on the Pareto frontier for video editing along with Fable, Grok, and open-source models. Next few wks will be interesting!
3
0
7
1,663