On September 26, 2026, TechCrunch reporter Dominic-Madori Davis published a first-person account of turning herself into a Synthesia digital twin — four avatar versions from one studio session, including an interactive one trained to answer questions about a single story. Synthesia says it was the first avatar it had built for a journalist, and it puts AI trust and disclosure squarely on the newsroom table.
2026-09-26 · by Moe Ameen
On September 26, 2026, TechCrunch reporter Dominic-Madori Davis published a first-person piece describing how she created an interactive AI avatar of herself using Synthesia, the London-based avatar-video company that hit a $4 billion valuation earlier this year and had crossed $100 million in ARR the year before. Synthesia says it was the first time it had built a digital avatar for a journalist.
The capture was fast and low-effort. Davis went into a small film studio inside Synthesia's office, where the team took a batch of photos and recorded roughly a two-minute sample of her voice. From that, Synthesia produced four versions: two "personal" avatars that read any supplied script aloud (one with glasses, one without) and two "interactive" avatars (again, with and without glasses) that can listen to a spoken question and respond in real time. The personal avatars are the familiar script-to-video product; the interactive ones are the newer capability.
The interactive avatar was deliberately fenced in. It was trained on Davis's own reporting about venture-backed startup fraud and will only answer questions about that story — ask it anything else and it steers back to the article. Under the hood, Synthesia describes an interactive avatar as a stack of voice-to-text, language, and text-to-voice models feeding its proprietary video model; customers can swap in third-party model providers such as Cartesia, ElevenLabs, Google, or OpenAI for parts of that chain. Synthesia's Interactive Avatars product line runs as real-time conversational agents that can join a live video session, with pay-as-you-go pricing published at $0.12 per minute.
Davis framed the experiment around trust rather than novelty. She noted that she had to consent to the avatars being made, and that the hardest part of journalism — a reader's trust that a real person did the reporting and stands behind it — is not something that can be handed to an AI. That is the open question the launch surfaces: as synthetic likenesses move into newsrooms and brand channels, audiences need to know when they are watching an avatar versus a human, which makes consent and clear disclosure the load-bearing safeguards.
The instinct after a story like this is "I should make an AI twin." The more useful read is what the reporter landed on: the value isn't the avatar, it's trust — a consistent identity you consented to and can disclose honestly. That is exactly the premise [Kompozy](/) is built on. Its avatar output — [Persona Shorts](/glossary/persona-shorts), Persona HeyGen video, and face-locked Persona Photos — is generated from an AI Influencer persona you set up and control, governed by a [Persona Brief](/glossary/persona-brief) so the voice stays yours across every post. You're never borrowing a stranger's likeness, so the consent question the Synthesia piece raises is answered by construction: the on-screen identity is one you own or one your team agreed to.
Where Synthesia's experiment ends at a single interactive twin, Kompozy's job starts at distribution. It's a generation and publishing engine, not a single-clip studio: from one source it produces recurring avatar video plus Clipped Shorts, Carousel Posts, Quote Graphics, Photo Posts, Blog Articles, and Email Newsletters, then [Autopilot](/glossary/autopilot) fans the batch across the eight social platforms plus your blog and a Mailchimp newsletter. And because nothing ships unattended — a per-post review pipeline sits in front of publishing — adding a clear "AI-generated" or synthetic-media disclosure to a caption or on-screen text is a deliberate step, not an afterthought. That's the honest, repeatable version of what this news is pointing at: an identity you consented to, disclosed plainly, published everywhere. For the legal and production side of likeness-based content, see [unauthorized AI likeness ads](/guides/unauthorized-ai-likeness-ads) and [where YouTube's AI disclosure and likeness rules stand in 2026](/news/youtube-ai-disclosure-likeness-rules-2026).
In a piece published September 26, 2026, TechCrunch reporter Dominic-Madori Davis created an AI avatar of herself with Synthesia. From a short studio session — a photo set and about a two-minute voice recording — Synthesia produced four versions: two "personal" avatars that read any script aloud and two "interactive" avatars that listen and respond in real time (each rendered with and without glasses). Synthesia says it was the first avatar it had made for a journalist.
No. The interactive avatar was trained specifically on Davis's reporting about venture-backed startup fraud and will only answer questions about that story — other questions get redirected back to the article. Scoping a conversational avatar to a defined knowledge base like this is a deliberate accuracy guardrail that stops the likeness from improvising answers it can't stand behind.
Because a convincing synthetic version of a real person can now be made from a two-minute sample, audiences increasingly can't tell an avatar from a human on sight. Davis's own conclusion is that trust — the core of journalism — can't be outsourced to an AI. The practical safeguards are explicit consent to the likeness (she had to consent to her avatars being made) and clearly labeling synthetic media so viewers know what they're watching.
The lesson isn't "make a twin," it's "own a consistent, consented identity and disclose it." A content engine like Kompozy generates avatar video from a persona you control (governed by a Persona Brief), produces clips, carousels, images, blogs, and newsletters from the same source, and publishes across nine platforms behind a per-post review gate — so you can add a synthetic-media disclosure before anything ships, then distribute everywhere.