
Descript Adds Persistent Custom Instructions to Underlord
Descript has added custom instructions to Underlord, its AI-powered editing assistant, allowing users to define their editing preferences once rather than repeating them in every session. The feature, announced on LinkedIn, addresses a common friction point in AI-assisted editing: the need to re-establish context and preferences each time you open a project.
Users can now set preferences such as pacing (faster cuts, specific transition styles), formatting choices, or any other recurring editing decisions. Underlord stores these instructions and applies them automatically to future requests, eliminating the need to describe the same style parameters repeatedly.
The feature is accessible through Settings → AI Models → Underlord Instructions.
How Custom Instructions Work
Custom instructions function as a persistent memory layer for Underlord. Once configured, they influence how the AI interprets and executes editing requests across all projects. This differs from the recently added chat history feature, which retains conversation context within individual projects but doesn't carry stylistic preferences across your entire workspace.
The implementation is straightforward: users describe their preferred editing approach in natural language, and Underlord incorporates these guidelines into its decision-making process. For example, specifying "keep cuts tight, under 2 seconds between scenes" or "always use cross-dissolve transitions between interviews" would inform how Underlord structures edits without requiring those instructions in each prompt.
This complements Underlord's existing context features. Users can still attach specific scenes, timestamps, or layers to individual requests using the @ syntax, but custom instructions provide baseline behavior that applies unless explicitly overridden.
Broader Context of Underlord Development
Descript has been rapidly iterating on Underlord since its initial release. Recent updates include support for Claude Opus 4.8 and Google Gemini 3.5 Flash, giving users the ability to select different AI models based on the complexity and precision requirements of specific editing tasks.
The company also ran a Telethon event in May 2026 focused on bug fixes and user-requested features, signaling a commitment to refining the editing workflow based on real-world usage patterns.
Custom instructions represent a logical next step in making AI editing assistants more practical for professional use. Rather than treating each session as isolated, the feature allows Underlord to function more like a trained collaborator who understands your house style.
Implications for Video Editing Workflows
For creators working on series content, branded videos, or client projects with established style guides, custom instructions reduce repetitive prompting and help maintain consistency across episodes or deliverables. This is particularly relevant for solo creators or small teams where maintaining stylistic coherence across multiple projects can be challenging without dedicated editors.
The feature also lowers the barrier to using AI assistance effectively. New users often struggle with prompt engineering; custom instructions let them establish working parameters once and refine them over time rather than optimizing every individual request.
Descript continues to position itself as a collaborative editing platform where AI augments rather than replaces editorial judgment. Custom instructions fit this model by allowing users to define boundaries and preferences while still maintaining control over specific creative decisions in real time.
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.




