ElevenLabs — speech recognition in every language, and the audio dub. Google Gemini — translation and hook ideas. Have I Been Pwned — the leaked-password check. Flitt — card payments. Each works under its own terms; the Privacy Policy says what each one gets.

Omnilingual ASR (CTC 1B, Meta) — Apache License 2.0. Run with PyTorch (BSD 3-Clause), Transformers (Hugging Face, Apache 2.0) and SentencePiece (Google, Apache 2.0).

The Georgian language model — built by us from Georgian Wikipedia (kawiki-20260901) by the Wikipedia contributors (ka.wikipedia.org), under CC BY-SA 4.0 and the GFDL. The model holds word counts and probabilities, not the text; because of ShareAlike its files are themselves CC BY-SA 4.0 when shared, which does not extend to the GeoCaptions program.

KenLM — Kenneth Heafield, GNU LGPL 2.1. pyctcdecode — Kensho, Apache 2.0. pygtrie — Google, Apache 2.0. Silero VAD (finding speech and pauses) — MIT License.

LR-ASD (who is speaking) — code and weights © 2025 Liao Junhua, MIT License. github.com/Junhua-Liao/LR-ASD

AVA-ActiveSpeaker — the data set LR-ASD's weights were trained on. © Google LLC, made available under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence. Roth et al., “AVA-ActiveSpeaker: An Audio-Visual Dataset for Active Speaker Detection”, research.google.com/ava.

MediaPipe BlazeFace (face detection, full-range model) — © Google LLC, Apache License 2.0; run with LiteRT (Google, Apache License 2.0).

PySceneDetect (shot cuts) — © Brandon Castellano, BSD 3-Clause. OpenCV — Apache 2.0. SciPy, NumPy — BSD 3-Clause. python_speech_features — MIT License.

FFmpeg — LGPL 2.1 or later; the server's build (with libx264) is GPL 2 or later. It runs as a separate program.

Pillow (drawing the captions) — MIT-CMU licence; with libraqm (MIT), FriBiDi (LGPL 2.1) and HarfBuzz (MIT) for right-to-left and joined scripts. fontTools — MIT License.

Flask, Werkzeug, Jinja, MarkupSafe — BSD 3-Clause. gunicorn — MIT License. yt-dlp (importing a video from a link) — The Unlicense; yt-dlp-ejs — The Unlicense (bundling astring, MIT, and meriyah, ISC); Deno (for YouTube links, run as a separate program) — MIT.

Paper Shaders (the home page's moving background) — © 2026 Paper, Apache License 2.0. shaders.paper.design

Noto Sans Georgian, Noto Serif Georgian, Noto Sans (Latin and Cyrillic), Noto Sans Armenian, Noto Sans Arabic, Noto Sans Hebrew — Copyright 2022 The Noto Project Authors, SIL Open Font License 1.1.

FiraGO — Copyright Carrois Corporate GbR and HERE Europe B.V. (Fira Sans: The Mozilla Foundation, Telefonica S.A., Carrois Corporate GbR and bBox Type GmbH). SIL Open Font License 1.1.

DejaVu Sans Bold (also the last-resort font for symbols such as → ★ ✓ ①) — Fonts are © Bitstream, Inc.; glyphs imported from Arev fonts are © Tavmjong Bah; DejaVu changes are in the public domain. The Bitstream Vera / DejaVu font licence.

Noto Color Emoji (Google) — SIL Open Font License 1.1. The server uses it only for an emoji your device did not draw.

Otherwise your own device draws the emoji with its own emoji font (for example Apple's on an iPhone or Mac), and those pictures go with your video only. The server keeps no copy of those fonts.

The speech in the example video every new account starts with is seven sentences from FLEURS (Google), read by volunteer speakers, under CC BY 4.0 (huggingface.co/datasets/google/fleurs). The moving background and the captions are ours.