Twenty-two themes, mixable
Pick a theme and stay in it, or hit Mix All and let a dragon wear a robot's antenna. Custom Mix narrows the pool to the themes you actually want in the shuffle.
Build a character.
Then let it speak.
Twenty-two themes — robots, dragons, sea creatures, anime, angels, flowers — layered part by part, and mixable across all of them at once. No account to create, nothing to install, no photo of you involved anywhere.
Your character is yours — nothing is published unless you share the code.
What you get
Every avatar is a short code
Your whole character — parts, colours, background, lighting — compresses to one share code. Paste it back any time to rebuild it exactly. Downloaded PNGs are named with their code, so an old file can always be reopened and edited.
No photo, no account
A character here is built, not scanned. There is nothing to upload, no sign-up to get through, and no picture of your face anywhere in the process — because there was never a picture of your face to begin with.
TarTalk — give it a voice
An avatar built here can talk. Type a line and the mouth follows the sound through fifteen distinct shapes, blended rather than snapped, so it reads as a jaw and not a flip-book. Download the result as an MP4 and post it anywhere.
Fifteen mouth positions
The standard animator's viseme set, driven from the audio waveform in real time — plus blinking, brow movement and a breathing head bob.
Records to video
Capture the performance straight to video and download it. Where the browser supports it that is an MP4 you can post as-is.
Nothing to set up
Ready-made voice clips and your system's own speech voices. No microphone permission, no sign-up, no install.
TarTalk Pro
in developmentEverything the public page deliberately holds back: real speech synthesis on anything you type, live microphone performance, musical timing, and a conversation mode where the character talks back.
Real voices, no limits
Neural text to speech on arbitrary text, in several voices and accents — plus a model that runs entirely inside your own browser, so long scripts cost nothing and go nowhere.
Reader
Hand it a poem or a transcript and it reads aloud at a considered pace, not the same delivery at a lower speed — line breaks land as pauses, punctuation buys real silence, and the mouth follows the phrasing instead of the syllable count.
Emote
Eyes and brows that carry an emotion and settle back to neutral, a gaze that mostly holds still and occasionally moves, and a register that follows what is being said rather than looping the same expression.
Dialogue
Push to talk and hold a conversation. It transcribes you, thinks, and answers out loud — remembering what was said — with a live transcript beside it and both halves recorded to one video.
Live microphone
Drive the mouth from your own voice in real time. The mic reaches the analyser and the recorder but never the speakers, so it cannot feed back.
Bring your own key
The model-backed features run on your provider account — Anthropic, OpenAI, Google, Groq, Mistral, OpenRouter. Nothing is resold and there is no shared quota to run out of.
Also being built, and not included in a plan yet: Rap, Sing. They will be part of Pro when they are good enough to charge for, not before.
Get told when it opens Not open to the public yet — no waiting list games, just one email when it is.
Start with a face
It takes about ten seconds to get something you like, and the code that rebuilds it is one line long.