invideo's Agentic Extend lengthens a scene by up to 30 seconds

invideo's editor will now make a scene longer because you asked it to. On October 6, 2026, invideo announced Agentic Extend, a feature inside invideo Editor that continues a shot by up to 30 seconds from a written request, generates the new material inside the project and leaves it on the same timeline you edit by hand.

What changed is where the work happens. Generative video tools have been good at producing a clip and clumsy at changing one you already have. Extending a finished shot has meant exporting its last frame, generating a separate clip somewhere else, then assembling the two pieces in an editor. invideo says Agentic Extend runs that loop inside the edit.

How Agentic Extend extends a scene

The feature is conversational. An editor brings filmed or generated footage, describes what should happen next and what should stay consistent, and the invideo agent uses the existing scene to guide the continuation. invideo says the agent analyses the scene's visuals and audio, then generates the new material within the project. Each extension runs up to 30 seconds, per the announcement.

A request can land in four places. The agent can generate beyond a shot's last frame, create a lead-in before its first frame, fill a missing or deleted section inside a scene, or regenerate a selected section in place. invideo's FAQ page repeats that list and says the added material can go before a shot, after it or into a gap in the sequence.

The audio continues on its own layers

Agentic Extend does not stop at picture. invideo says it can continue a cut-off line in the same voice, extend music and carry sound effects into the new footage, holding the scene's characters, products and audio identity as the action develops.

Voice, music and sound effects stay on separate layers. An editor can shift their timing, change the balance or replace one element without rebuilding the whole sound bed. One limit is worth stating plainly. A continued line of dialogue is generated speech, not a recovery of what the performer originally said, invideo notes on the feature.

Where the editor still has to be the editor

The generation lands on the same multitrack timeline the editor controls. There the editor can inspect the new material, trim the continuation, refine the audio layers and send the agent back for further changes.

invideo puts the split this way: the agent carries out the generation, and the editor decides which result serves the scene and how it fits the finished cut. That framing is the maker's, and it sets the feature against a version of AI editing where the tool returns a locked clip. Here the output is editable by design.

The workflow also accepts outside agents. Through invideo MCP, an editor can connect ChatGPT or Claude, or clients including Codex, Cursor, VS Code and Gemini CLI, to the same workspace. The agent opens a project, imports media and edits the timeline while the editor watches, working under its own cursor. Both can edit at the same time. From there a project exports as a video or as a project file for Premiere Pro, Final Cut Pro or DaVinci Resolve.

What it costs

AI generation in invideo Editor uses invideo credits. Manual timeline editing is free and costs no credits, per invideo's help center, which says trimming, splitting and rearranging clips carries no charge.

invideo's credits page says credits are priced by the model, the output length and the resolution, and that unused plan credits do not roll over between billing cycles. The credit cost is shown before each generation is confirmed. invideo's help center adds that video models are priced per second.

How it compares to Adobe and Runway

Two shipping rivals mark the gap this closes. Adobe's Generative Extend, generally available in Premiere Pro since April 2, 2025, adds up to two seconds of video and ten seconds of audio to a clip that is at least two seconds long, by dragging its edge on the timeline. Adobe's documentation says the generated frames are labeled and that the feature runs on Firefly and uses generative credits.

Runway's approach differs again. In its help docs for Seedance 2.5, Runway's Extend mode continues an existing video forwards or backwards from a prompt, and every extension comes back as a standalone clip. The output holds only the new footage, the original is unchanged, and the editor assembles the pieces in Studio. A single Seedance 2.5 generation can run up to 30 seconds.

invideo's version extends by conversation rather than by dragging an edge, and it continues the audio on separate layers rather than returning picture alone. It also generates inside the project, where Adobe works through the timeline and Runway hands back separate clips to reassemble.

Conversational editing is not itself new. Descript has offered Underlord, an AI co-editor that works from a chat sidebar. The new part in invideo's release is extending picture and carrying a scene's dialogue, music and sound effects through the same generated pass.

What invideo has said and what comes next

invideo's announcement keeps returning to one claim: the editor stays the decision-maker. The agent generates the material, the review is manual, and the sequence can be adjusted at any point. That is the maker describing its own product, and the verification for it will come from editors who run Agentic Extend on a real cut.

The feature does not arrive with a public list of which models power the continuation, how long the generated material stays consistent, or how many extensions a single shot supports.

Agentic Extend is available now in invideo Editor. AI generation uses invideo credits, manual editing is free, and agents connect through invideo MCP at https://invideo.io/mcp.