Soku AI
Edit Video Ads by Editing the Transcript
Descript turns a rough recording into a clean, captioned cut without a timeline. Soku AI has no native Descript connection — export the cut and upload it to Soku to build the variants and tests a paid campaign needs.
Can I connect Descript to Soku AI today?
Not yet availableNot yet. Soku AI does not have a native Descript connection, so it can't read, sync, or act on Descript data on its own. This page explains what Descript does and how it fits next to Soku — it is not a description of a live integration.
Request Descript
Tell us you need it via the contact form or the Feedback button inside the Soku app. Requests are tracked on the product roadmap.
Bring your own MCP server
Organization admins can connect any tool that offers a remote MCP server in Settings → Integrations → Custom external source, using an API key or OAuth. Custom sources are enabled per organization — ask us to turn them on for your workspace.
Descript at a Glance
Descript is an audio and video editor built around the transcript: delete a sentence and the footage goes with it, move a paragraph and the video follows. On top of that sits Underlord, an AI co-editor with 20+ actions, plus recording, captions, translation, and dubbing.
Core Capabilities
Transcript-Based Editing
The edit surface is the transcript. Deleting a sentence removes the footage for that sentence; rearranging paragraphs rearranges the cut. Anyone who can edit a document can cut a video ad.
This is why non-editors on a marketing team can ship their own creative refreshes.
Underlord AI Co-Editor
An AI layer with 20+ actions: strip filler words and dead air, generate captions, propose short clips from a long recording, tighten pacing, and write titles and descriptions.
Full Underlord access starts on the Creator plan; Free and Hobbyist are limited.
Studio Sound
Cleans up audio recorded on a laptop mic or a phone — removes room tone and background noise and brings the voice forward, so a founder recording at a desk sounds close to studio-recorded.
AI Captions
Burned-in captions generated from the transcript, styled per brand. Since most paid social is watched muted, this is a conversion feature, not a cosmetic one.
Translation & Dubbing
Produce localized versions of one hero video without a re-shoot — translated captions plus AI dubbing, so a single asset can seed several market tests.
Available on Business and Enterprise plans.
Screen & Remote Recording
Record your screen, your camera, or a remote guest in separate tracks, straight into the project — no separate capture tool in the chain.
Public API
Import media from a URL or upload, run Underlord actions, edit a project's composition, and export via a shareable link or download URL. Tokens are generated in Settings and scoped to a Drive.
Early access. API usage is included for paying plans and draws on existing AI credits and media minutes.
MCP Server
Descript ships an MCP server, so an AI assistant can drive an edit directly instead of a human clicking through the UI.
From Raw Recording to Tested Ad in 4 Steps
Descript owns the cut. Soku owns the campaign. Neither tool tries to do the other's job, which is why the handoff is clean.
Record and Clean
Record a walkthrough, a founder take, or a customer story. Studio Sound fixes the audio and Underlord strips filler words — no re-take, no audio engineer.
Cut by Editing Text
Trim a two-minute take to 25 seconds by deleting sentences in the transcript. Add captions from the same transcript. Export at 1080p or 4K, watermark-free.
Expand into Variants with Soku
Soku takes the cut and generates what a real test needs — 9:16 and 1:1 crops, alternate opening hooks, alternate end cards — instead of one asset per placement by hand.
Launch, Measure, Re-Cut
Soku runs the variants as a structured test across Meta, Google, and TikTok, reports which hook carried the spend, and tells you exactly which section to re-cut next time.
Built for Video-Led Performance Teams
Founder-Led Talking-Head Ads
Record once at a desk, clean it with Studio Sound, cut by deleting sentences — then let Soku turn one take into the eight variants a creative test actually requires.
Long-Form to Paid Social
A webinar or podcast becomes 20–40 second clips. Underlord proposes the clips; Soku decides which ones deserve budget and which are just interesting.
Multi-Market Rollouts
Translation and dubbing produce localized cuts of one hero video. Soku runs each locale as its own test instead of assuming one winner travels.
Creative Fatigue Refreshes
When performance decays, the fix is usually a new first three seconds — a five-minute transcript edit, once Soku has told you which creative fatigued and when.
Customer Story Ads
Turn a recorded customer call into a testimonial spot: cut the transcript to the strongest 30 seconds, caption it, and test it against your existing hooks.
Agency Client Throughput
Account teams cut their own video without booking an editor, and Soku keeps every client's variants, tests, and results separated by brand.
How Soku AI and Descript Fit Together
Descript makes one asset good. Soku makes it a campaign — variants, placements, tests, and the performance read that tells you what to cut next.
Cut Once, Test Many
A single Descript export becomes a structured creative test in Soku: platform-native crops, alternate hooks, and end-card variants launched together rather than one at a time.
The variant set is generated from the asset, so the test is comparable by construction.
Performance Points Back at the Transcript
Because Soku tracks which variant and which hook carried the spend, the next Descript session starts from evidence — keep the opening that worked, re-cut the section that lost attention.
Automation Is Not Native Yet
Descript exposes a public API and an MCP server, but Soku has no native Descript connection, so the record → clean → cut → test loop is not driven end-to-end today. Organizations with custom sources enabled can try connecting a remote Descript MCP server as a Custom external source.
Descript's API is early access — keep a human on the export.
Descript Pricing
Plans scale on media hours and AI credits. API usage is included for paying plans and draws from the same pool rather than adding a separate bill.
Plans
| Plan | Annual / Monthly | Included |
|---|---|---|
| Free | $0 | 1 hour of media/mo, 100 one-time AI credits, 720p export with watermark, limited Underlord |
| Hobbyist | $16 / $24 per user/mo | 10 hours/mo, 400 AI credits/mo, watermark-free 1080p, unlimited projects |
| Creator | $24 / $35 per user/mo | 30 hours/mo, 800 AI credits/mo, 4K export, full Underlord access, unlimited stock media |
| Business | $50 / $65 per user/mo | 40 hours/mo, 1,500 AI credits/mo, team of 5, Brand Studio, translation & dubbing, custom avatars |
| Enterprise | Custom | Custom media hours and credits, SSO, SCIM, audit logs, AI data controls, dedicated CSM |
What This Means for Ad Teams
| Question | Answer |
|---|---|
| Cheapest plan with full Underlord | Creator — Free and Hobbyist cap the AI co-editor |
| Cheapest watermark-free export | Hobbyist ($16/user/mo annual, 1080p) |
| Plan needed for dubbing | Business or Enterprise |
| Extra cost for API access | None — included on paying plans, draws on plan credits |
How Descript Compares
Descript's differentiator is the editing model itself: the transcript is the timeline. The trade-off is that it is built for dialogue-led content, not cut-to-music montage.
| Capability | Descript | CapCut | Adobe Premiere | Riverside |
|---|---|---|---|---|
| Editing Model | Transcript-driven | Timeline + templates | Professional timeline | Transcript + timeline |
| Learning Curve | Low — edit like a document | Low | High | Low |
| AI Editing Layer | Underlord, 20+ actions | AI templates & effects | Firefly / Sensei tools | AI clip generation |
| Audio Cleanup | Studio Sound | Basic noise reduction | Essential Sound / Enhance | Magic Audio |
| Translation & Dubbing | Business plan and up | Limited | Via add-ons | Yes |
| Public API | Early access, included in plan | No public editing API | Extensibility via CEP/UXP | Limited |
| MCP Server | Yes | No | No | No |
| Free Tier | 1 hour/mo, watermarked | Generous free tier | Trial only | Limited free tier |
| Starting Paid Price | $16/user/mo (annual) | ~$8/mo | ~$23/mo | $15/mo (annual) |
Frequently Asked Questions
What is Descript?
Descript is a video and audio editor built around the transcript. Instead of scrubbing a timeline, you edit text: deleting a sentence removes the corresponding footage, and rearranging paragraphs rearranges the cut. It also includes screen and remote recording, an AI co-editor called Underlord, Studio Sound audio cleanup, AI captions, and translation and dubbing.
Can I connect Descript to Soku AI?
Not natively — there is no Descript connection in Soku today. Descript produces the finished cut; Soku turns it into a campaign. You export a clean, captioned asset from Descript, and Soku generates the platform-specific variants a real creative test needs — 9:16 and 1:1 crops, alternate hooks, alternate end cards — launches them across Meta, Google, and TikTok, and reports which variant carried the spend so you know what to re-cut next.
Does Descript have an API?
Yes, in early access. The public API covers importing media from a URL or upload, running Underlord actions such as filler-word removal, captions, clip generation and Studio Sound, editing a project's composition, and exporting via a shareable link or download URL. Tokens are created under Settings → API Tokens and are scoped to a specific Drive.
Does API access cost extra?
No. Descript states that API access is available to all paying users at no additional cost, and that usage draws from the AI credits and media minutes already included in your plan rather than being billed separately.
Is there a free plan?
Yes. The free plan includes 1 hour of media per month and 100 one-time AI credits, with 720p exports that carry a watermark and limited Underlord access. It is enough to evaluate the editing model, but not enough to run a paid-social workflow — Creator is the first plan with full Underlord access and 4K, watermark-free export.
Which Descript plan should an ad team choose?
Creator at $24/user/mo billed annually is usually the honest starting point: it is the first tier with full Underlord access, 4K watermark-free export, and 30 media hours a month. Move to Business if you need team collaboration, Brand Studio, or translation and dubbing for multi-market ad variants.
Can Descript replace a traditional video editor?
For dialogue-led content — founder takes, talking-head ads, webinars, podcasts, customer stories — largely yes, and much faster. For cut-to-music montage work where the edit is driven by beats rather than words, a conventional timeline is still the better tool. Most performance teams use Descript for the first category and treat the second as a separate production track.
How do Descript and Soku divide the work?
Descript makes one asset good. Soku decides what happens to it: how many variants to generate, which placements to run, how to structure the test, and which creative earns more budget. Descript has no campaign layer and Soku is not a timeline editor, so the handoff is a clean boundary rather than an overlap.
Want Descript in Soku AI?
A native Descript integration is not available yet. Tell us you need it — requests shape what we build next.