Underlord is Descript's AI editing assistant, introduced as a way to make video production more conversational. Rather than forcing creators to learn every editing command, Underlord can interpret instructions and make changes to the project using the transcript and media context.

What Descript does

Descript says Underlord can edit by acting on the transcript and can handle tasks including filler removal, clip selection, show notes, and more complex editing requests.

The system can also work with visual context. Descript has described engineering work that allows Underlord to identify moments based on what happens on screen rather than relying only on spoken words.

Key capabilities to know

  • Descript says Underlord can edit by acting on the transcript and can handle tasks including filler removal, clip selection, show notes, and more complex editing requests.
  • The system can also work with visual context. Descript has described engineering work that allows Underlord to identify moments based on what happens on screen rather than relying only on spoken words.
  • Descript has introduced a model picker so users can choose among supported AI models for different tasks.
  • Underlord can assist with video generation, captions, layouts, audio improvements, thumbnails, and other production steps.
  • The platform remains a full editor, so creators can manually inspect and refine AI-generated changes.

How the workflow works

The practical advantage of an AI editor is not that every decision becomes automatic. It is that repetitive operations can be expressed as intent: remove pauses, tighten a section, create a clip, or prepare a publishable version.

That makes Underlord particularly relevant to content teams producing many variations from the same source recording.

  1. Start with a recording, transcript, or existing project.
  2. Describe the desired edit in plain language and provide any important constraints.
  3. Review the proposed changes and inspect the exact media segments affected.
  4. Manually refine the result, then create captions, clips, thumbnails, translations, or exports.

Practical use cases

  • Turning webinars into multiple social clips.
  • Removing filler words and retakes from interviews.
  • Creating a first cut from a scripted recording.
  • Preparing translated or repurposed versions of existing videos.
  • Generating supporting assets such as show notes and thumbnails.

What to consider before adopting it

Natural-language editing works best when prompts contain concrete goals and constraints. A vague instruction can produce an acceptable but creatively wrong result, so review remains important for professional publishing.

Model selection also introduces a quality-versus-cost tradeoff. Different tasks can benefit from different models, particularly when a job involves complex planning or heavy media reasoning.

Bottom line

Underlord makes Descript more than a transcription editor: it turns the transcript, timeline, and media assets into a workspace that an AI assistant can operate. That can materially reduce repetitive production work without removing the human editor from the process.