claude-real-video lets Claude watch a video from the terminal. Kompozy generates and publishes content across 9 platforms — no CLI. The honest 2026 comparison.
If you searched "claude-real-video alternative," it is worth being honest about what you actually found. claude-real-video is a free, open-source command-line tool that lets Claude — or any LLM — watch a video: it extracts scene-change frames, transcribes the audio, and hands the model a manifest so Claude can answer questions grounded in what is on screen. It is a genuinely clever piece of engineering, and if analyzing a video from the terminal is your goal, it does that well. This page is not going to pretend otherwise.
I run Kompozy, and the honest framing is that these two tools sit on opposite sides of the workflow. claude-real-video is an input tool — it makes a video legible to a model so you can understand it. Kompozy is an output engine — it generates finished content and publishes it across nine platforms. They barely overlap. So most people who search this "alternative" are not really looking for a second way to parse video for an LLM; they got to a terminal utility, realized it makes nothing you can post, and started looking for the tool that actually produces content.
That is the useful fork. If you need to interrogate a video — mine a competitor's clip, understand a long course, run private local analysis — claude-real-video is the right shape and you do not need an alternative. If your gap is that you have footage and ideas but no captioned shorts, carousels, blog, or scheduled posts to show for them, an analysis CLI is the wrong shape for that job, and that is what people are really shopping for here.
Everything below reflects claude-real-video's state as of 2026-07-02, read from its public repository: an MIT-licensed, Python-based, local CLI for turning video into frames plus a transcript. No invented weaknesses.
claude-real-video (run as the command crv) is a local command-line tool that gives Claude or any LLM the ability to watch a video. Claude cannot ingest a raw video file, so the tool does the perception step: it uses ffmpeg to detect scene changes and capture a frame at each meaningful transition, deduplicates near-identical frames against a sliding window, and transcribes the audio — preferring an existing subtitle track and falling back to Whisper. It can pull a video straight from a YouTube, Instagram, or TikTok URL via yt-dlp, or work on a local file, and it can optionally preserve the full soundtrack. The result is a folder of JPEG frames, a transcript, and a MANIFEST file the model reads. Everything runs on your machine with nothing uploaded to a cloud service, and it is free and MIT-licensed. It is a perception layer, not a content workflow. It does not generate captions, cut a clip for you, reframe video for vertical, write a post or a blog, produce a carousel or an image, enforce a brand voice, or publish anything anywhere. It makes a video readable to Claude; what you do with that understanding — and everything you might want to create and post — is a separate job entirely.
The reason to look past claude-real-video is not quality — it is that it is not a content tool and never claimed to be. It is a developer utility you run from a terminal, which means Python, ffmpeg, and command-line flags before you get a single frame. There is nothing to publish at the end: the output is frames and text meant for a model to read, not posts meant for an audience. There is no clip export, no branded captioning, no per-platform reframing, no scheduling, and no posting to any network. There is also everything a piece of video can become that an analysis CLI has no path to: a set of vertical Clipped Shorts, a Carousel of the key points, quote graphics, native Text Posts per platform, a recap Blog Article, an Email Newsletter, and net-new formats like a HeyGen avatar Persona Short. None of this makes claude-real-video weak — it makes it an input tool that still needs a generation-and-publishing engine downstream if your goal is to turn video and ideas into content that ships.
| Feature | claude-real-video | Kompozy | Note |
|---|---|---|---|
| Let an LLM "watch" a video (frames + transcript) | Yes — the core strength | No | claude-real-video is purpose-built to make a video legible to Claude. Kompozy generates content; it is not a model-analysis tool. |
| Scene-change frame extraction | Yes | Partial | crv extracts analysis frames for a model. Kompozy detects moments to cut Clipped Shorts, a different goal. |
| Local, private, no-cloud processing | Yes | No | crv runs entirely on your machine. Kompozy is a cloud content engine. |
| Free / open-source (MIT) | Yes | No | crv is free and open-source. Kompozy is a paid, hosted product. |
| Runs without a terminal / for non-developers | No — CLI only | Yes | crv needs Python, ffmpeg, and command-line flags. Kompozy is a web app. |
| Cut vertical shorts from long-form video | No | Yes | Kompozy Clipped Shorts detects strong moments and cuts them to vertical with branded captions. |
| Auto-captions / branded caption styling | No | Yes | Kompozy burns in branded captions via HyperFrames; crv outputs raw frames and text. |
| Generate carousels, quote cards, images | No | Yes | Kompozy produces carousels, quote graphics, and infographics; crv generates nothing. |
| Write blogs and newsletters | No | Yes | Kompozy drafts a recap blog and newsletter governed by the Persona Brief. |
| Brand voice / Persona Brief governance | No | Yes | Kompozy enforces tone and banned phrases per workspace across every output. |
| Multi-platform publishing | No | Yes | Kompozy schedules and fans to 9 platforms + blog + email from one queue; crv publishes nothing. |
| Cost model | Free (MIT, self-run) | Monthly credits | crv is free to run yourself; Kompozy bills monthly credits covering generation + publishing. |
| Tier | claude-real-video plan | claude-real-video price | Kompozy plan | Kompozy price |
|---|---|---|---|---|
| Entry | claude-real-video (self-run) | Free (open-source, MIT) | Kompozy Creator | $49/mo (2,500 credits) |
| Mid | claude-real-video + your own LLM/API cost | Free tool; you pay any model/API usage | Kompozy Pro | $299/mo (18,000 credits) |
| Top | claude-real-video (unchanged) | Free — no paid tier exists | Kompozy Enterprise | Custom (sales-led) |
Here is the honest pitch. claude-real-video is a good tool for one specific job — making a video watchable by Claude — and if that is your whole goal, use it and skip everything else; it is free and it works. But understanding a video is the start of a content workflow, not the end of one. The frames and transcript it produces are for a model to read, not an audience to watch, and nothing in that pipeline cuts a clip, writes a caption, or posts anything anywhere.
Kompozy is the engine on the other side of that gap. If you want the outcome — content, on brand, across platforms — you do not start in a terminal. Bring your long-form video into Kompozy and Clipped Shorts detects the strongest moments, cuts them to vertical, and burns in branded captions through HyperFrames. The same session becomes a Carousel of the key points, quote cards, native Text Posts, a recap Blog Article, and an Email Newsletter — all in your voice through a Persona Brief — and Kompozy generates net-new formats a transcript can't, like HeyGen avatar Persona Shorts. Then Autopilot schedules and publishes the whole set across all nine connected platforms plus your blog from one queue.
You can even run both: analyze a competitor's video or your own back catalog with claude-real-video to decide what to make, then let Kompozy make and ship it. But if all you actually want is content, the analysis CLI was never the tool — Kompozy is. Start on Kompozy Creator at $49/mo (2,500 credits).
Not exactly — they do opposite jobs and can pair. claude-real-video is an input tool that lets Claude watch and analyze a video from the terminal. Kompozy is an output engine that generates finished content — shorts, carousels, blogs, newsletters — and publishes it across nine platforms. If your goal is to make and post content rather than analyze a video, Kompozy is the tool; if you specifically need a model to watch a video, keep claude-real-video.
No. It extracts frames and a transcript so an LLM can understand a video, but it does not cut clips, caption them, reframe them, or publish anything. To turn a video into scheduled posts across platforms, bring it into a content engine like Kompozy.
Effectively yes. It is a Python command-line tool that depends on ffmpeg and yt-dlp and is driven by flags, so you need to be comfortable in a terminal. Kompozy, by contrast, is a web app with no command line, Python, or ffmpeg to install.
Cut long-form video into captioned vertical shorts, generate carousels, quote graphics, blogs, and newsletters, produce HeyGen avatar video, enforce brand voice with a Persona Brief, and schedule and publish across nine platforms plus a blog and email. claude-real-video only prepares frames and a transcript for a model to read.
claude-real-video is free and open-source under an MIT license — you run it yourself and pay only for whatever model reads its output. Kompozy is a paid, hosted product starting at $49/mo (2,500 credits) that covers content generation across formats plus multi-platform publishing.