How Video Editors Use Descript Underlord for Rough Cuts and Social Clips

Video editors working with long-form recordings—such as podcast interviews, webinars, and unscripted talking-head footage—traditionally lose hours to the rough cut phase. Scrubbing through two hours of raw multicam footage just to find false starts, stumbled lines, and five-second soundbites is exhausting and repetitive.

Descript addresses this bottleneck through Underlord, an integrated AI co-editor accessible directly from the sidebar. Instead of forcing editors to cut manually along a complex audio-video timeline, Underlord couples transcript-driven text editing with natural language prompts. Editors can purge failed takes, remove filler speech, assemble narrative rough cuts, and slice long files into vertical social media clips in a fraction of the time.

Here is the exact practical workflow video editors use to go from raw, unedited footage to polished rough cuts and published social clips using Descript Underlord.

1. Ingestion and Speaker Identification

Before running any AI commands, start with accurate media ingestion:

1. Import Raw Media: Drop your multitrack video files, Zoom recordings, or raw camera audio directly into Descript.
2. Generate Transcript and Identify Speakers: Descript automatically creates an interactive transcript synchronized to every frame of the timeline. Immediately use the Identify Speakers tool to label every host and guest.

Why this matters for Underlord: Tagging speakers allows Underlord to understand conversational dynamics. When you later request multi-speaker cuts, automated multicam switching, or speaker-focused vertical framing, the AI relies on these tags to make smart editorial decisions.

2. Generating the Rough Cut with Underlord

The traditional rough cut requires deleting bad takes, cleaning up speech stumbles, and shortening silent gaps. Underlord handles these housekeeping edits through a combination of dedicated tools and conversational prompts.

Step A: Purge False Starts with "Remove Retakes"

Creators frequently re-record a sentence two or three times before getting the delivery right.

• Open the Underlord sidebar (or click the sparkle AI Tools icon).
• Run Remove Retakes (or prompt Underlord: "Remove all retakes and keep the best take").
• How it works: Underlord analyzes sentence patterns in the transcript to detect repeated phrases or failed attempts. It automatically keeps the final, complete sentence and hides the failed attempts.
• The Non-Destructive Advantage: Descript marks rejected takes as Ignored text rather than permanently deleting media. If Underlord misjudges an intentional repetition or comedic pause, you can restore that section instantly by selecting the struck-through text and hitting restore.

Step B: Eliminate Filler Words and Tighten Cadence

• In Underlord, trigger Remove filler words. The tool highlights common filler speech ("um," "uh," "you know," "like") across all tracks. You can review them in bulk or strip them in one click.
• Run Shorten word gaps to eliminate awkward dead air between sentences (e.g., capping pauses at 0.5 to 0.8 seconds).

Step C: Apply Studio Sound for Clear Audio Assessment

Poor room acoustics and volume discrepancies make it difficult to evaluate the pacing of an interview. Ask Underlord to apply Studio Sound across speaker tracks. This strips background hiss, HVAC noise, and room echo, creating an isolated, broadcast-grade audio profile before you make narrative decisions.

Step D: Conversational Story Trimming

For interviews with rambling tangents, you can direct Underlord using conversational instructions in the chat box:

• "Create a tight 8-minute rough cut focusing exclusively on the product design debate, cutting introductory small talk and tangents."
• "Edit for clarity and tighten pacing without changing speaker intent."

Underlord modifies the text transcript, which immediately ripples across the underlying video timeline.

3. Repurposing Long-Form Content into Social Clips

Once the core long-form recording is structured, editors turn Underlord toward social media repurposing for platforms like TikTok, Instagram Reels, YouTube Shorts, and LinkedIn.

Step A: Automated Highlight Extraction

Instead of re-listening to the entire conversation to note timestamps:

1. Open the Underlord sidebar and select Create clips under the Repurpose section (or type: "Find 5 high-energy highlights suitable for TikTok").
2. Set your parameters:
   • Number of clips: Between 1 and 20.
   • Clip length: Target duration between 15 seconds and 2 minutes.
   • Focus or Topic: Optional prompts (e.g., "Focus on controversial opinions about remote work").
3. Click Submit. Underlord analyzes speech density, engagement hooks, and narrative closure, generating distinct sub-compositions for each clip.

Step B: Converting to 9:16 Vertical Layout & Auto-Multicam

Social clips demand vertical (9:16) presentation:

• Switch the composition aspect ratio to 9:16 Portrait.
• For two-person interviews recorded in 16:9 widescreen, apply Automatic Multicam or speaker tracking. Underlord detects who is actively speaking and pans or switches cameras automatically, keeping the speaker centered in the vertical frame.

Step C: Dynamic Animated Captions

Because over 70% of mobile users watch short-form video on mute, captions are mandatory:

• Select the clip text and choose Captions from the Underlord panel.
• Choose dynamic animated presets (e.g., word-by-word karaoke highlights, bold typography, custom brand hex colors).
• Reposition caption blocks within the mobile "safe zone" to avoid overlapping UI buttons on TikTok and Reels.

Step D: Brand Consistency via Custom Instructions

To avoid reconfiguring font styles, caption colors, and pacing rules for every project, navigate to Settings > AI Models to set Custom Instructions:

• Provide ongoing editorial rules: "Always generate vertical 9:16 clips with Montserrat Bold captions, yellow word highlights, and cross-zoom transitions on scene changes."
• Underlord remembers these preferences across sessions, standardizing output across an editing team.

4. Comparing Descript Underlord Workflow with Traditional NLEs

5. Pricing Plans and AI Credit Considerations

Descript uses a hybrid pricing model that balances monthly media minutes (footage ingestion) with AI credits (operations powered by Underlord, Studio Sound, and voice cloning).

Credit Management Tip: Because complex, open-ended Underlord prompts consume more credits during their reasoning steps, use targeted tools (like dedicated "Remove Retakes" and "Remove filler words" buttons) for mechanical housekeeping, reserving conversational AI prompts for substantive structural trimming and highlight curation.

6. Practical Editorial Best Practices

• Duplicate before Prompting: Descript's text-based editing is flexible, but running extensive conversational trimming prompts can significantly alter your project layout. Always duplicate your main composition before asking Underlord to make major structural revisions.
• Treat Underlord as a Junior Assistant: Underlord excels at repetitive manual labor, but it does not have human editorial taste. Review cuts where irony, comedic timing, or rhetorical pauses were used, as automated algorithms might misinterpret them as hesitations.
• Leverage the Hybrid Pro NLE Pipeline: For high-end client deliverables, use Descript Underlord for the first 80% of the project (ingest, retake removal, rough cut assembly, and social clip discovery), then use Export > Timeline (FCPXML or Premiere XML) to finish color grading and mastering in DaVinci Resolve or Premiere Pro.

References

• Underlord | Your AI Video & Podcast Editing Assistant | Descript
• AI Remove Retakes | Edit out bad takes with a click
• AI Video Clip Maker & Repurposing Tool | Descript
• Descript Pricing | Plans for Every Creator, Free to Start
• Descript Help Center