llm-real-video Pro

Purchased before July 14? Your download predates the license key file — just re-download the latest version from your Capafy orders page to get your LICENSE-KEY.txt. No extra cost.

Your LLM reads subtitles, not videos. crv Pro turns any video into evidence an LLM can reason about — what was said, what was shown, and how it was made.

Watch any video. Keep only what matters.

Point it at a YouTube/Instagram/TikTok URL or a local file. Scene-aware extraction keeps a frame only where the picture actually changes: a 58-second clip keeps 26 frames instead of 58. One user benchmarked an 8.8-minute launch video — 20 frames instead of 527 at 1 fps, with the whole story intact. Contact sheets pack the survivors so the reading cost drops about 5x.

It reads the film language, not just the frames.

Every shot is labelled — static, pan, tilt, zoom, handheld — with per-shot duration, cuts per minute, and how the pacing shifts across the open, middle and close. Voice emotion and tone curves are tracked segment by segment; vocals are auto-separated from the music, which gets its own BPM and energy line. Music videos and film footage with no dialogue switch to reading mood from colour and light.

Know who's talking — and how they sound.

Interviews, podcasts and meetings finally read like conversations: every transcript line carries a speaker label, fully local, no account or token needed. Combined with the voice-emotion track, the transcript tells you not just what was said, but who said it and how it landed.

From raw footage to a written report, in one flag.

--ai-report writes two reports on the spot with your own API key — one on how the video is shot, one on what it says (both together cost well under $0.10). Every analysis also ships with an interactive viewer: the video, a clickable event timeline that jumps to the second, and a transcript that highlights along with playback. Interface in English, Traditional and Simplified Chinese.

Trusted where it counts.

Everything runs on your machine — client footage and unreleased cuts never leave it. The frame timestamps have been independently verified by a user who replayed a 22-minute lecture with SHA-256 checks (public GitHub issue). And when the reading LLM has web access, the built-in source-metadata block guides it to verify who appears, where it was filmed, and whether the title matches the visuals.

In the wild

A creator pulls apart why a viral reel holds attention. An editor deconstructs a reference cut shot by shot. A music producer reads the BPM, energy and mood of an MV. A researcher fact-checks footage with timestamped receipts. A field engineer turns a site-visit video into an illustrated report. If you feed video to Claude, ChatGPT, Gemini or a local model, this is the missing eye.

How it ships

A Claude Code skill: paste a video link in chat and the agent watches it for you. Your download contains the installer (.whl) and your license key file (LICENSE-KEY.txt) — activation is fully offline, nothing phones home. One-time $29. Yours to keep, includes updates. Purchase terms: https://leoaido.com/crv-pro/terms/

Requirements

Python 3.10+, ffmpeg. macOS / Linux / Windows (Windows users may prefer WSL). Install guide included (INSTALL.md).

Support

Email support included. Issues are typically answered within 24 hours. If crv Pro helps your workflow, a rating on Capafy helps other creators find it — and honest feedback shapes what gets built next.