Repository navigation
Releases
selfclone 0.2.0: explainer videos starring you
Turn one photo of yourself into a cartoon host, and let Claude make explainer videos on any topic starring you. Everything renders on your own computer, on the CPU: no GPU, no image model and no video model.
showreel.mp4
Highlights
- Explainer videos on any topic. Ask Claude for a video about anything. It checks the facts, writes the script and a scene file, previews stills with storyboard warnings, and renders an MP4 with captions. The 30 to 40 second showcase films each rendered in 20 seconds to 2 minutes on a laptop CPU.
- You as the host.
selfclone character --photo me.jpg --consent "..."draws a cartoon of you from one photo or a short clip. Or cast random fictional people (random:<seed>), or put nobody on screen (--no-host). - Characters that act. Jointed arms with gestures chosen from the words, six moods, glances at each new item, head turns, blinks, a walk cycle, and lip sync read from the phonemes and timed to the voice. In a dialogue each character has its own voice.
- Voices. Kokoro on the CPU by default (pinned and hash-checked), ElevenLabs, OpenAI, or your own recording.
- Everything a lesson needs. 60 props, bar, line and pie charts that grow, handwritten text, bubbles, counters, your own images, drawn scenery with parallax, camera moves and transitions, sound effects and a generated music bed. Manim for maths, LaTeX included.
- Made to post. 16:9, 9:16 and 1:1 frames, outlined captions plus an
.srt, and an interactive HTML player (--html). No watermark. - Ten showcase films in
examples/showcase, one per field: AI, aerospace, finance, medicine, maths, English, neurology, philosophy, physics (as a 9:16 short) and astronomy. - Also new: Voice CI (lint a draft against your own writing habits and a project's rules), the Voice Passport (a public card of your writing style with no message text), voice drift, and
selfclone video, which turns the photo avatar into a picture-in-picture presenter lip-synced to your own recorded voice.
Install
In Claude Code:
/plugin marketplace add logitsolutionsai/selfclone
/plugin install selfclone@selfclone
Then ask: "make me a character from me.jpg and an explainer video about how vaccines work".
You need Node 20 or newer, ffmpeg, and Chrome, Edge or Chromium. Manim is optional, for maths. Read ETHICS.md before using anyone's face but your own.
Full list of changes: CHANGELOG.md.
selfclone 0.1.0
First public release.
Give Claude your own chat exports and it builds a skill that writes like you. Record a 10-second clip, or take one selfie, and the clone gets a face that talks and blinks in a single offline web page. Then a score shows how close it gets, measured against real messages of yours that were sealed away before the clone was built.
Highlights
selfclone ingestreads WhatsApp (Android and iOS), Telegram, Slack, X archives, mbox and email, plain notes, and JSONL or CSV. A scrubber removes keys, passwords, card numbers, phone numbers and emails before anything else runs.- A deterministic style score (function words plus 26 style features, calibrated on your own replies). In the shipped example the clone scores 89% and the same model with no clone skill scores 41%.
- A "Can You Tell?" game page: friends pick the real reply out of two.
- The face: MediaPipe finds the face once at enrol time (about 7 MB, cached), then a standalone
avatar.htmlwarps your photo in WebGL with lip sync, blinks and small head movement. No model weights, no cloud service, no video export. - Consent is built in: the face comes from your own webcam or a file you confirm is your own, the consent record is baked into the page, and an "AI CLONE" badge is always on.
selfclone serveandselfclone saylet a Claude Code session talk through the face.
Install
In Claude Code:
/plugin marketplace add logitsolutionsai/selfclone
/plugin install selfclone@selfclone
Or see the README for a plain skill install and the command-line tool. Read ETHICS.md before cloning anyone other than yourself.
Full list of changes: CHANGELOG.md.