Skip to main content
Hashnode
MirrorCut BlogMirrorCut Blog

7 Best Opus Clip Alternatives in 2026 (Open-Source and Hosted)

Updated 10 min readMarkdown
M
MirrorCut

Every other tool here is described from its own site, repo or launch post, checked Oct 30, 2026. Please check them again before you decide anything.

October was the month of "I built my own Opus Clip".

  • Danish Sofi wrote up Hive Clips AI, a nine-stage clipper he built with Claude Code.
  • SupoClip (open source, AGPL-3.0) launched on Product Hunt on Oct 14.
  • Katto lets you clip from inside Claude, ChatGPT, Gemini CLI or Cursor.
  • CutVii shipped as a free, open-source AI editor.

The reasons are familiar: subscriptions add up, credits run out, and you can't always tell a black box what you want. Opus Clip is still good at what it does, and it now takes natural-language prompts through ClipAnything. But if you're shopping for an alternative, open-source or hosted, here are seven worth a look, and how to choose.

What matters in an Opus Clip alternative

  • Selection you can steer. "Every time they talk about pricing", not just "the most viral 30 seconds".
  • Reframing that doesn't shake. Detecting a face is easy; a smooth virtual camera is the hard part.
  • Captions that break lines in the right place and stay in sync after edits.
  • A human review step before anything renders or posts.
  • What happens to the long video. Most clippers assume it's already clean. For raw talking-head recordings, it usually isn't (more on that below).
  • Cost and licence: credits vs subscription vs your own compute, and what an open-source licence asks of you.

1. MirrorCut: clean up the long recording, then clip it

MirrorCut is an AI agent that edits your recordings in the browser. It's built for raw talking-head recordings, where you want the long video and the clips.

  • What the AI does: removes dead air, filler words, retakes and false starts in the long recording first. Then it finds Shorts, scored by hook, each on its own timeline so you can edit it separately. Captions (48 styles) stay in sync after cuts, and reframing to 9:16, 1:1 or 4:5 follows the speaker.
  • Review and undo: every AI change is shown with a one-line reason and its own Accept/Revert.
  • Output: MP4 up to 4K, plus SRT and VTT.
  • Price: Creator $10/month, Pro $30, Studio $60; every plan has every feature (pricing).
  • Best for: solo creators whose long recordings need cleaning before they're worth clipping.

2. SupoClip: open source, self-host or hosted

SupoClip takes a long video (a podcast, a talk, a stream VOD), finds and scores moments, and renders vertical 9:16 clips with face-centred cropping, word-synced subtitles, hook titles and optional B-roll.

  • Ways to use it: a hosted web app, an iOS app, self-hosting with Docker Compose, an MCP server, and a REST API (per its README).
  • Licence: AGPL-3.0. If you modify it and offer it to others over a network, the AGPL's source-sharing terms apply. Read the licence before you fork it into a product (this is a general note, not legal advice).
  • Best for: people who want an Opus-style pipeline they can inspect and run on their own hardware.

3. Build your own (the Hive Clips-style pipeline)

Danish's write-up is the clearest public breakdown of what a clipper actually is:

Stage What happens Typical open-source tools
Ingest Get the long video, normalize it FFmpeg
Transcribe Words with timestamps Whisper / faster-whisper
Select Find candidate moments An LLM with structured JSON output
Review A human picks which to keep Your UI
Configure Aspect ratio, caption style Your UI
Reframe Track the speaker, move a virtual camera OpenCV, PySceneDetect
Caption Word-by-word animated captions ASS subtitles, libass
Render Crop, burn, encode FFmpeg
Deliver Files, links, uploads Your storage

Most of it is free software. The paid part is usually the model calls for transcription and selection, and Danish notes you can run those locally to get close to zero.

  • Where the time goes: selection and reframing. Naive face tracking shakes and slides across camera cuts "like a drunk camera operator"; his fix is a critically damped spring plus a deadzone, and he still lists cut handling as unfinished. Captions aren't trivial either: a recent DEV post shows auto captions breaking English lines in the wrong place and fixes it with spaCy (post).
  • Price: your compute, API calls and time.
  • Best for: client work with custom briefs, footage that can't leave your machines, high volume, or learning.

4. Katto: clip from inside your AI assistant

Katto turns a long video into short vertical clips, finding the moments worth posting, and lets you run it from Claude, ChatGPT, Gemini CLI or Cursor.

  • Good at: fitting into the chat or coding tool you already use.
  • Price: see Katto's site.
  • Best for: people who'd rather ask their assistant for clips than open another app.

5. CutVii: a free, open-source AI editor

CutVii is a new free, open-source editor (launch post Oct 24) with AI auto clips, auto subtitles, AI voice-over, LUTs, a timeline and 4K export, per its author.

  • Good at: a full editor plus clipping, at no cost.
  • Watch out for: it's brand new, so expect rough edges and check the repo's activity.
  • Best for: tinkerers who want an open-source editor, not just a clipper.

6. Descript: transcript editing plus clips

Descript edits video like a doc. Its chat co-editor Underlord handles Remove Filler Words, Remove Retakes, captions and clips.

  • Review and undo: Underlord creates a checkpoint you can revert to; per Descript's help docs, checkpoints last only for the current session (help).
  • Hand-off: timeline export to Premiere, Final Cut Pro and others.
  • Price: Free (60 media minutes a month, 720p export); Hobbyist $24/month or $16 billed annually (pricing).
  • Best for: clipping and cleanup when you finish in an NLE.

7. Riverside: record, then Magic Clips

Riverside is a recording studio first, with editing on top: Magic Clips for short clips, an AI Co-Creator, and silence and filler-word removal.

  • Price: Pro $29/month, or $24/month billed annually (pricing).
  • Best for: podcasters who record remote guests in Riverside and want clips from the same place.

Side-by-side

Tool Open source / self-host Cleans the long video first Clips Cost to start*
MirrorCut No / No Yes (dead air, filler, retakes, false starts) Shorts scored by hook $10/mo
SupoClip Yes (AGPL-3.0) / Yes Not its focus Scored 9:16 clips Free self-host; hosted plans on site
Build your own Yours / Yes Only if you build it Your rules Compute + API
Katto See site Not its focus Vertical clips from your assistant See site
CutVii Yes / See repo Editor timeline AI auto clips Free
Descript No / No Yes (filler words, retakes) Clips via Underlord Free; $24/mo ($16 annual)
Riverside No / No Silence and filler removal Magic Clips $29/mo ($24 annual)

*Checked Oct 30, 2026, from each tool's own pages. Prices change, so click through first.

The step most clippers skip

Here's the part I care about, because it's where I spend my days.

Clippers assume the long video is already good. For a polished podcast with an editor, it is. For a solo creator's raw talking-head recording, it usually isn't. The raw take has:

  • the three seconds of silence before you start
  • the line you said twice because you fluffed it
  • the "um, so, basically" at the top of every answer
  • the false start you abandoned halfway

A clipper picks a 40-second moment from that. If the best moment contains a retake, the clip contains the retake, and you're back in an editor fixing it by hand. A clean long video is better to publish and better to clip from.

Which one should you pick?

Your situation Sensible pick
Raw talking-head recordings, want the long video and clips MirrorCut (#1) or Descript (#6)
Want Opus-style clipping you can inspect or self-host SupoClip (#2)
Custom selection rules, client batches, privacy needs Build your own (#3)
You live in Claude, ChatGPT or Cursor Katto (#4)
Free, open-source editor to tinker with CutVii (#5)
You finish in Premiere or Final Cut Descript (#6)
You record remote guests Riverside (#7)
Polished podcast, clips only, budget matters Opus Clip itself is still fine (Free: 60 credits with a watermark; Starter $15/month)

If you build: defaults worth stealing

  • Word-level timestamps from day one. Every later stage needs them.
  • Overlapping transcript windows for selection, so a great moment isn't split across two LLM calls. Then merge and re-rank.
  • A strict JSON schema for the model's output, with fixed min/max clip lengths. Danish asks for 2–5 clips of 20–90 seconds per window.
  • Persist every stage. A failed render should re-run only the render.
  • A deadzone and smoothing on the crop, never raw per-frame face centring.
  • A mock mode that skips paid API calls, so you can test reframing and rendering for free.

And what you sign up to maintain: edge cases forever (hard cuts, two speakers in frame, music under speech), a job queue and storage, licences, model drift (keep a small test set of videos with known good clips), and rights. Clip footage you own or have permission to use.

Before you choose: a checklist

  1. Write down the three clip briefs you'd actually give it. If "most viral moment" is enough, you may not need to switch at all.
  2. Collect five real recordings, including a messy one, as a test set.
  3. Run the same set through two or three tools above and judge the picks.
  4. Decide where the human review step lives. Danish's "nothing renders until you pick" was, in his words, the thing he missed most in paid tools.

If you've built or switched clippers this month, I'd like to hear which stage surprised you most. My bet is reframing. And if you want to try the clean-up-first approach, MirrorCut is at mirrorcut.ai.

Sources (checked Oct 30, 2026)

Comments

Join the discussion

No comments yet. Be the first to comment.

More from this blog