Higgsfield ships Katana, an AI video editor that runs inside Claude via MCP

What Katana does inside Claude

Higgsfield introduced Katana, powered by Claude Motion, and pitched it as an AI video editing tool that lives inside Claude rather than as a standalone desktop application. The product connects through the Higgsfield MCP, so a user opens Claude, attaches the connector, uploads a reference visual, and prompts the model to produce an editable project from that reference. Posts published on October 11, 2026 describe the workflow as upload-then-prompt, with creators sharing motion design and character work generated by Katana in response to a single prompt.

5, used to drive code-based motion design paired with Katana output.

Reference-driven edits from a creator's own footage

A second demonstration thread tests Katana with personal footage rather than text-only prompts. The creator uploaded their own robot photos and clips and walked through the Katana in Claude using the Higgsfield MCP, asking the system to produce a robot reveal. The result is positioned as a practical test: can the reference-to-edit pipeline recreate a specific look from a creator's own library, and can an editor steer the output by adjusting the prompt inside Claude rather than rebuilding the edit in a traditional timeline.

Other October 11 posts frame the same capability more broadly, calling Katana "an absolute banger for making videos" and describing it as motion design that Claude can now do on the editor's behalf.

What changes for an editor's workflow

The shift for practitioners is the location of the editing surface. The tool is no longer a panel inside Premiere Pro or After Effects, nor a standalone NLE, but a connector inside a chat assistant that returns an editable project from a reference. That has three concrete consequences for video creators and editors.

First, the upload that feeds the edit can be any visual, including photos lifted from an existing cut, which Extract Frames from Video: How to Pick the Right Time and a tool such as Video Frame Extractor make easier to prepare. Second, the reference frames themselves have to be picked deliberately, since How Do I Extract a Video Frame at a Specific Time is the kind of upstream step that now feeds the prompt.

Third, because Katana returns an editable project rather than a finished render, the downstream export still goes through a normal encoder, where Why Can the Output Be Larger With Video Compression explains why a re-encode can blow up the file size if the settings are wrong. The October 11 chatter frames the release in combative language, with one caption declaring that Katana has "killed After Effects," which signals how disruptive creators perceive the move to an in-chat editing model.

What to verify next

Two open questions will determine whether Katana becomes a daily tool or a novelty. First, what file formats and timelines Katana exports, since that controls whether the output drops cleanly into existing Premiere Pro, After Effects, or DaVinci Resolve projects. Second, how the reference upload is compressed and stored, since a Video to MP3 in Your Browser: What You Actually Get style comparison is the kind of practical read an editor will want before trusting the connector with source footage.

Editors who want to try the workflow without committing source material can begin by extracting stills on device with Extract Frames from Video on Android Without Uploading and feeding those stills as the reference visual. The Higgsfield MCP listing inside Claude is the public entry point for that test on October 11, 2026.