Descript Adds Claude Fable 5 to Underlord for Video-Aware AI Editing.

Descript has added Anthropic's Claude Fable 5 to its Underlord AI assistant, bringing video-aware capabilities that let the model analyze footage and automatically optimize common editing decisions like zoom, framing, and on-screen text.

The update, announced on LinkedIn, positions Underlord to handle more of the visual decision-making traditionally left to editors. According to Descript, the model can identify movement during screen shares and apply appropriate zoom, refine text overlays, frame speakers automatically, and handle credit attribution for AI-generated elements.

Underlord launched earlier this year as an AI-powered editing assistant inside Descript. The tool operates through natural language prompts, allowing users to request edits, adjustments, or workflow automations without manually navigating the interface. Recent updates introduced memory and custom instructions, enabling Underlord to remember user preferences across projects.

Claude Fable 5 extends that by adding vision. The model can parse what's happening on screen, identify key elements, and apply edits that typically require manual scrubbing and layer adjustments. For creators working with screen recordings or talking-head videos, this could compress several steps into a single prompt.

Descript's LinkedIn post highlights four specific capabilities:

• Detecting and zooming into active elements during screen shares
• Refining on-screen text overlays for clarity or positioning
• Automatically framing speakers based on their position in the frame
• Burning AI credits into the final output

The company hasn't published detailed documentation yet, but the announcement suggests these functions are available now inside Underlord for users with access.

Underlord has steadily gained model options and workflow features since its debut. Descript previously added Claude Opus 4.8 for more deliberate, step-by-step edits, and Google Gemini 3.5 Flash for speed-focused tasks. The integration of multiple models gives users flexibility depending on the task: quick cuts, thoughtful revisions, or now, video-aware optimizations.

The addition of vision capabilities aligns with broader industry movement toward multimodal AI. Tools like Claude and Gemini have been rolling out image and video understanding across their platforms, and Descript is applying that directly to the editing workflow.

For creators managing screen recordings, tutorial videos, or presentations, the ability to prompt zoom adjustments or speaker framing without manually keyframing could meaningfully reduce production time. The feature is especially relevant for solo creators or small teams where editing bandwidth is limited.

Descript's announcement was brief, leaving several operational questions unanswered. It's not yet clear how granular the controls are, whether users can override or refine the model's suggestions inline, or how credit burning works in practice. There's also no indication of whether Claude Fable 5 is available across all Descript plans or limited to higher tiers.

The company has been shipping updates rapidly in recent months, often through internal hackathons and customer-driven feature requests. According to its June changelog, Descript ran a 48-hour telethon in which the team built 70 customer-requested features. Many of those updates are still rolling out, suggesting this release is part of a broader product acceleration.

Underlord continues to differentiate Descript from traditional video editors by shifting editing from a timeline-first model to a prompt-first one. With Claude Fable 5, the assistant can now see what it's editing, which may further compress the gap between intent and output.