Meta hired celebrities to voice AI characters, then moved on
Meta's 2023 launch names celebrity-embodied AI characters; its 2024 follow-up shifts the product to user-made ones instead.
Read piece →Topic shelf · 18 records
Voice & avatars
Embodiment, synthetic presence, and the mechanics behind a character surface.
This shelf follows the extra product constraints introduced by speech, timing, synthetic identity, animation, and visual embodiment.
OpenAI paused a ChatGPT voice after a likeness complaint
OpenAI's own account says it paused the Sky voice on 19 May 2024 after Scarlett Johansson objected, without confirming imitation.
Read piece →GPT-4o's voice launch came with a warning about emotional reliance
OpenAI's own announcement and system card, plus reporting on the September rollout, show what the company flagged about voice and attachment.
Read piece →Replika's paywall sits on relationship status and voice
Replika's help pages and store listing set out what Free Use, Pro, Ultra and Platinum unlock, as retrieved on 16 September 2026.
Read piece →Hume's voice interface says it measures expression, not emotion
Hume's own documentation describes prosody measurement and an empathic model, not detection of felt emotion.
Read piece →Sesame published its method for a voice that feels present
Sesame's research post describes a model built for voice presence and commits key components to an open license.
Read piece →ElevenLabs builds agent voices on a self-attested consent claim
ElevenLabs' documentation and terms describe agent voices, cloning, and the consent language that governs them.
Read piece →Meta gave its assistant five actors' licensed voices, not clones
Meta's announcement named five actors whose AI voices became selectable, distinct from a cloned voice.
Read piece →Replika layered its avatar onto a room through AR and VR headsets
Replika's own help pages describe AR and headset modes as beta features layered onto a text-based companion.
Read piece →VRM and Live2D encode different rules for who owns a digital face
The VRM specification and Live2D's licenses show how differently each format records who may use an avatar.
Read piece →A voice companion needs a text door, a repeat button and time to answer
Accessibility testing should cover modality switching, transcripts, correction and pacing—not just whether speech recognition works in a quiet room.
Read piece →Measure a voice companion’s pauses before calling it natural
A phone, a spreadsheet and a fixed script can separate network delay, response generation and speech playback without pretending to be a laboratory.
Read piece →Every borrowed face and voice needs a consent ledger
Creators need evidence of who authorized an identity, for which uses, for how long, and what happens when permission ends.
Read piece →Design a calm mode for people who need less companion at once
Reduced motion, literal language, predictable controls, adjustable pacing and reversible steps make companionship more accessible without making it cold.
Read piece →Character Calls turned character chat into a live turn-taking surface
The 2024 rollout joined voice selection, interruption, and text switching into one interaction loop.
Read piece →Character.AI’s June 2025 tools moved the unit of creation beyond chat
Scenes, animation, profiles, and feeds suggest a platform mechanics shift: characters become reusable media objects.
Read piece →
Vizlings puts 3D embodiment and casts ahead of a single chatbot thread
The developer’s App Store listing describes animated characters, voice and text, creation tools, worlds, and group interaction.
Read piece →
Animates’ stated approach makes animation a local companion primitive
Animation Inc. describes real-time generative animation, while its companion page foregrounds a character’s ongoing presence.
Read piece →