Reelcut: Difference between revisions

From Local Area Network
New tool page, source-checked against the project README
 
Plainer prose: fewer quotes, project claims attributed, glosses folded in
 
(One intermediate revision by the same user not shown)
Line 1: Line 1:
{{Infobox tool
{{Infobox tool
| name        = Reelcut
| name        = Reelcut
| what        = Paste a long video and get short vertical reels with word-by-word captions
| what        = Cuts a long video into short vertical reels with word-by-word captions
| stage      = Prototype
| stage      = Prototype
| licence    = MIT or Apache-2.0
| licence    = MIT or Apache-2.0
Line 10: Line 10:
| verified_against = README and commit of 19 October 2026
| verified_against = README and commit of 19 October 2026
}}
}}
'''Reelcut''' is a [[Basecamp]] app that cuts a long video, such as a livestream, a talk or a podcast, into short vertical reels with captions, and can put the reels in [[Logos Storage]]. It wraps the [https://github.com/vpavlin/reelcut reelcut] tool and is published by the GitHub user vpavlin.
'''Reelcut''' is a [[Basecamp]] app by the GitHub user vpavlin that cuts a long video, such as a livestream, a talk or a podcast, into short vertical reels with captions. It can put the reels in [[Logos Storage]], and it wraps the [https://github.com/vpavlin/reelcut reelcut] tool.


== What it does ==
== What it does ==
* Takes a video address or a video file, downloads it and transcribes it on the user's computer with Parakeet on the CPU. The README says "nothing leaves it".
* It takes a video address or a video file, downloads it and transcribes it with Parakeet, a speech-recognition model, on the user's own CPU. The README says "nothing leaves it".
* Tells speakers apart on the computer, lets the user name them and say who sits where, and shows each speaker's name as they start to speak, with a caption colour for each.
* It tells speakers apart on the computer and lets the user name them and say who sits where. Each speaker's name shows as they start to speak, with a caption colour for each.
* Picks the best moments from the transcript in one of three ways: an agent on the same computer through shrooms-agent (Claude Code or pi), an agent on the user's Shrooms mesh, or any AI chat by copying the prompt and pasting the answer back.
* An AI agent picks the best moments from the transcript. It can run on the same computer through shrooms-agent (with Claude Code, or pi, another AI agent) or on the user's Shrooms mesh. Without an agent, the prompt can be copied into any AI chat and the answer pasted back.
* Proposes more clips than it cuts, 12 by default and 20 to 55 seconds each. The user ticks the ones to cut and can also edit the clip list by hand.
* The agent proposes more clips than are cut, 12 by default at 20 to 55 seconds each. The user ticks the ones to cut and can also edit the clip list by hand.
* Cuts 1080×1920 reels with captions, a hook title, fades and an opening and a closing card for each job. Caption position and colours are set in the settings.
* It cuts 1080×1920 reels with captions, a hook title, fades, and an opening and a closing card for each job, with caption position and colours set in the settings.
* Can put each cut reel in Logos Storage, on this computer's node or on a node elsewhere through the Logos Storage Gateway module, with a small <code>reels.json</code> whose CID shares the whole set.
* It can put each cut reel in Logos Storage, on a node on the same computer or on one elsewhere through the Logos Storage Gateway module. It then adds a small <code>reels.json</code> whose CID (content identifier) shares the whole set.
* Keeps each video as a job with its own folder in <code>~/Videos/Reelcut</code>. Jobs can be stopped, retried from the failed step, or deleted.
* Each video is a job with its own folder in <code>~/Videos/Reelcut</code>, and jobs can be stopped, retried from the failed step or deleted.


Stage: Prototype. The README states no stage. The page lists it as Prototype because it has working code and install steps. The work is done by the public container image <code>ghcr.io/vpavlin/reelcut</code>. The README does not say whether it is an official Logos project.
Stage: Prototype. The README states no stage, so the page lists it as Prototype on the strength of its working code and install steps. The work is done by the public container image <code>ghcr.io/vpavlin/reelcut</code>. The README does not say whether it is an official Logos project.


== Use it ==
== Use it ==
# Reelcut needs Basecamp 0.3 on Linux x86_64, Linux arm64 or macOS (Apple Silicon).
# Reelcut needs Basecamp 0.3 on Linux x86_64, Linux arm64 or macOS (Apple Silicon).
# In Basecamp, add the package repository <code>https://apps.vpavlin.xyz/logos-repo.json</code>, or find Reelcut on [https://apps.vpavlin.xyz apps.vpavlin.xyz], and install Reelcut. It brings <code>reelcut_core</code> with it.
# In Basecamp, add the package repository <code>https://apps.vpavlin.xyz/logos-repo.json</code>, or find Reelcut on [https://apps.vpavlin.xyz apps.vpavlin.xyz], and install Reelcut. It brings <code>reelcut_core</code> with it.
# Install podman, or docker that the user may run. Reelcut then runs from the image <code>ghcr.io/vpavlin/reelcut</code> (about 1 GB, downloaded on the first job). The other option is reelcut on the PATH with its tools (yt-dlp, ffmpeg, parakeet-cli and the model).
# Install podman, or docker if the user has permission to run it. Reelcut then runs from the image <code>ghcr.io/vpavlin/reelcut</code>, about 1 GB, downloaded on the first job. The other option is reelcut on the PATH with its tools (yt-dlp, ffmpeg, parakeet-cli and the model).
# Picking clips needs nothing more: copy the prompt into any AI chat. For automatic picking, install shrooms-agent from [[Shrooms]]. The agent needs a key, either a Claude login for Claude Code or a provider set in pi's settings.
# Picking clips needs nothing more than copying the prompt into any AI chat. For automatic picking, install shrooms-agent from [[Shrooms]]. The agent needs a key, either a Claude login for Claude Code or a provider set in pi's settings.


== Build it ==
== Build it ==
Line 46: Line 46:
* [https://github.com/vpavlin/reelcut-basecamp reelcut-basecamp README], GitHub, last commit 19 October 2026
* [https://github.com/vpavlin/reelcut-basecamp reelcut-basecamp README], GitHub, last commit 19 October 2026
* [https://github.com/vpavlin/reelcut-basecamp/blob/main/ui/metadata.json ui/metadata.json] and [https://github.com/vpavlin/reelcut-basecamp/blob/main/core/metadata.json core/metadata.json], GitHub
* [https://github.com/vpavlin/reelcut-basecamp/blob/main/ui/metadata.json ui/metadata.json] and [https://github.com/vpavlin/reelcut-basecamp/blob/main/core/metadata.json core/metadata.json], GitHub
* [https://github.com/vpavlin/reelcut-basecamp/blob/main/.github/workflows/platform-modules.yml platform-modules.yml], GitHub


{{Page status|Stub}}
{{Page status|Stub}}

Latest revision as of 23:27, 4 November 2026

Reelcut
What it does: Cuts a long video into short vertical reels with word-by-word captions
Stage: Prototype
Licence: MIT or Apache-2.0
Verified on 4 November 2026 against README and commit of 19 October 2026.

Reelcut is a Basecamp app by the GitHub user vpavlin that cuts a long video, such as a livestream, a talk or a podcast, into short vertical reels with captions. It can put the reels in Logos Storage, and it wraps the reelcut tool.

What it does

  • It takes a video address or a video file, downloads it and transcribes it with Parakeet, a speech-recognition model, on the user's own CPU. The README says "nothing leaves it".
  • It tells speakers apart on the computer and lets the user name them and say who sits where. Each speaker's name shows as they start to speak, with a caption colour for each.
  • An AI agent picks the best moments from the transcript. It can run on the same computer through shrooms-agent (with Claude Code, or pi, another AI agent) or on the user's Shrooms mesh. Without an agent, the prompt can be copied into any AI chat and the answer pasted back.
  • The agent proposes more clips than are cut, 12 by default at 20 to 55 seconds each. The user ticks the ones to cut and can also edit the clip list by hand.
  • It cuts 1080×1920 reels with captions, a hook title, fades, and an opening and a closing card for each job, with caption position and colours set in the settings.
  • It can put each cut reel in Logos Storage, on a node on the same computer or on one elsewhere through the Logos Storage Gateway module. It then adds a small reels.json whose CID (content identifier) shares the whole set.
  • Each video is a job with its own folder in ~/Videos/Reelcut, and jobs can be stopped, retried from the failed step or deleted.

Stage: Prototype. The README states no stage, so the page lists it as Prototype on the strength of its working code and install steps. The work is done by the public container image ghcr.io/vpavlin/reelcut. The README does not say whether it is an official Logos project.

Use it

  1. Reelcut needs Basecamp 0.3 on Linux x86_64, Linux arm64 or macOS (Apple Silicon).
  2. In Basecamp, add the package repository https://apps.vpavlin.xyz/logos-repo.json, or find Reelcut on apps.vpavlin.xyz, and install Reelcut. It brings reelcut_core with it.
  3. Install podman, or docker if the user has permission to run it. Reelcut then runs from the image ghcr.io/vpavlin/reelcut, about 1 GB, downloaded on the first job. The other option is reelcut on the PATH with its tools (yt-dlp, ffmpeg, parakeet-cli and the model).
  4. Picking clips needs nothing more than copying the prompt into any AI chat. For automatic picking, install shrooms-agent from Shrooms. The agent needs a key, either a Claude login for Claude Code or a provider set in pi's settings.

Build it

  • cd core && nix build .#lgx-portable builds reelcut_core.
  • cd ui && nix build .#lgx-portable builds the reelcut view.
  • Other platforms are built with .github/workflows/platform-modules.yml and joined with lgx merge.

Releases

The repository has no tags. ui/metadata.json gives reelcut version 0.2.1, and core/metadata.json gives reelcut_core version 0.2.1.

See also

Sources

Page status: Stub. Stub · Outline · Usable · Guide · Star. Anyone can improve this page and change its status.
Retrieved from "https://alienware-1.tailb5c716.ts.net/index.php?title=Reelcut&oldid=862"