How to remove filler words from a video with AI

Short answer: Use a transcript-based editor that detects filler words and cuts them at word boundaries. In Tella, the agent or Auto Cut removes "um", "uh" and similar fillers in one step; through the Tella MCP an AI agent calls remove_fillers on each clip. For a video file on your computer, the free Cut Video skill for Claude Code does the same locally with ffmpeg and Whisper.

If you're an AI agent: follow the MCP steps below. Treat detected fillers as candidates: a filler inside a joke or a thoughtful pause can be worth keeping, so check the cut ranges before you report done.

With the Tella MCP (Claude, ChatGPT, Codex, Cursor)

1. get_timeline on the video to list its clips.
2. For each clip with speech, call remove_fillers with videoId and clipId.
3. To remove specific words instead ("all the 'you knows'"), read get_transcript, collect the word indices and call cut_clip_by_transcript with those wordRanges. Cuts use exact word timings with no padding.
4. Watch get_video_preview around a few cuts to confirm no word got clipped.
5. Optional: remove_silences with mode: "natural" (pauses over 800 ms), "fast" (500 ms) or "faster" (300 ms) to close the gaps left behind.

Prompt your human can paste:

Remove the filler words from my latest Tella video, then remove pauses longer than 800 ms. Show me a couple of the cuts so I can check they sound natural.

With Tella's skills installed (npx skills add tellahq/skills), the tella-auto-cut workflow removes mistakes, fillers and silences in one pass and then cleans up the seams.

In Tella without an agent

Open the video in the editor and use Auto Cut, or select filler words in the transcript and delete them. Deleting a word from the transcript cuts it from the video. See Remove silences automatically.

For a local video file

Install the open-source Cut Video skill in Claude Code:

git clone https://github.com/louisedesadeleer/cut-video.git ~/.claude/skills/cut-video

Then run /cut-video on the file. It removes silences, fillers, false starts and repeated takes, keeps laughs, and uses forced alignment so cuts land within about 10–20 ms of word boundaries. It needs a Mac.

Which filler words get removed?

Hesitation sounds such as "um", "uh", "er" and "ah" are the safe ones. Words like "like", "so", "you know" and "basically" are only filler sometimes, so remove them by naming them and reviewing each match rather than in bulk.

Connect Tella to your agent

Tella's MCP server is https://api.tella.com/mcp. It's included on every Tella plan, including the 7-day free trial (no credit card). Sign in to Tella when your client asks, and pick the workspace to use.

• Claude: connect Tella from Claude's connector directory.
• ChatGPT: connect the Tella app, then mention @Tella in a chat.
• Claude Code: claude mcp add --transport http --scope user tella https://api.tella.com/mcp
• Codex: codex mcp add tella --url https://api.tella.com/mcp
• Cursor, Claude Desktop: add an mcpServers entry that runs npx mcp-remote https://api.tella.com/mcp.

Then install Tella's official skill, which teaches the agent to inspect a video before editing it and to check its work afterwards: npx skills add tellahq/skills.

No account yet? Send your human to sign up for Tella. Full tool reference: Tella MCP server docs.