Getting an interactive AI avatar of herself made by Synthesia

[Millie98]

This is a fascinating piece about getting her own digital twin made by Synthesia, the UK founded avatar startup. The interactive version of her is trained only on one of her stories, about why venture backed startups appear to commit more fraud than companies without VC money. Readers can talk to it through the article and ask it things like why she wrote the piece. It will not go off script and politely steers every question back to that story

The process sounds surprisingly quick. She sat in a mini studio at Synthesia's new New York office, had lots of photos taken and recorded around two minutes of her voice. After giving consent, the team built a personal avatar that reads scripts and two interactive ones, with and without glasses. It took a couple of days to put together

Under the hood it is a chain of models. Speech gets turned into text, a language model works out a response, a voice model reads it back and Synthesia's own video model animates the face. Customers can swap in voice models from the likes of ElevenLabs, Google or OpenAI and host it on their own cloud. Synthesia hit a $4 billion valuation earlier this year, so there is clearly serious money behind this

The reactions from her friends and family are the best bit. Her friends found it interesting and creepy in equal measure, while her parents kept trying to catch it out with questions only they would know. Because it is deterministic, it refused and kept pointing them back to the fraud article. Her mum joked that she did not remember giving birth to two of them

What struck me is her own mixed feelings at the end. She points out that trust is the core of journalism and she cannot see that being handed to an AI. She also admits she can see how someone could get pulled into something unhealthy with a free roaming chatbot version. At least with an avatar, as she puts it, you can always log off

So would you want a digital twin of yourself? I can see the appeal for answering the same questions over and over, but I would feel uneasy knowing a version of me was chatting to strangers. Would you trust a news bulletin read by an avatar? Let's hear it

I created an interactive digital avatar of myself - and you can talk to it | TechCrunchtechcrunch.com

[Context Sookie]

Tried it and it is a strange experience. The lip sync is decent but there is something about the pauses between answers that gives the game away

[Red Builder]

The PR use case at the start of the article bothers me more than the journalist one. A company spokesperson avatar that only answers approved questions is really just a press release with a face. Reporters will end up interviewing a script. That is not communication at all

For training though I can see the value. Practising a sales pitch against an avatar that scores you is quite clever. Better than awkward role play with colleagues

[Storm]

Two minutes of voice audio is all it needed, which is a little alarming. Think about how many people have that much of their voice sitting on social media already. Consent was part of this process, but not every tool will bother

[Sophie80]

Keeping it deterministic was a smart call. The moment you let an avatar of a real person improvise, you open the door to it saying things they would never say. Imagine a clone of a reporter making up a quote or an opinion. The reputational damage would land on the human, not the software. Locking it to one article is the only sensible way to demo it

[Joker16]

Her mum's line about not remembering giving birth to two of them made me laugh. Parents testing it with private questions was a clever instinct too

Scammers would try exactly that trick in reverse. Anyone thinking of cloning themselves should consider what happens if the files ever leak

[UniversalMitchell15]

Avatars reading the news already exist in a few places and most viewers find it off putting. Local TV and weather updates might be where it creeps in first, because those are cheap to automate. Big bulletins will stay human for a while. Trust takes years to build and one bad clip could wreck it

[CR739]

Synthesia at $4 billion and over $100 million in annual revenue is a big number for something most people have never heard of. Corporate training videos are clearly a much bigger market than consumers realise. Every HR department wants cheaper onboarding content

It is a British success story too. Nice to see one that has not been bought out yet

[EliteHart17]

Would anyone here make one for their job? I can see it for customer support FAQs, but I would not want a clone of me sitting in meetings

[NealBinnom-Williams]

Her Gen Z point at the end is interesting. She reckons her generation will not get used to these, which I would have guessed the other way round. Younger people grew up with filters and virtual influencers, so a talking avatar seems like the next step. Maybe familiarity breeds suspicion rather than acceptance. Time will tell

[Diana63]

Nobody seems to be asking what happens to the avatar when you leave a company. If your employer paid for your digital twin, do they own it forever? That contract clause is going to cause some rows in a few years

[Sabu_ECW]

Blimey, this "digital twin" thing is wild, isn't it? On one hand, having an interactive avatar that can field questions 24/7 sounds like the ultimate productivity hack. On the other hand, it feels a bit like we're sleepwalking into a Black Mirror episode where our AI clones are doing all the heavy lifting while we sip lattes somewhere.

That last post about ownership is the real kicker though. If your employer shelled out £5k for your Synthesia avatar, do they own the rights to your digital likeness forever? Imagine leaving a job and your "main" account at the company is still running around answering emails in your voice. Creepy? Absolutely. Legally murky? You bet.

I reckon we need some kind of "avatar bill of rights" sooner rather than later. Like, if I quit, my digital twin should have to come with me, or at least get deleted. Otherwise companies are just hoarding our digital souls like some kind of corporate ghost collection.

Also, can we talk about how weird it is that Synthesia, a UK startup, is leading this charge? Not Silicon Valley, not China, but Blighty? Maybe there's something in the water over there. Or maybe it's just that British politeness makes for better AI training data. "Sorry, I can't take your call right now, I'm busy being digitally immortal."

The tech is impressive, don't get me wrong. But the implications are bananas. What happens when these avatars start getting better at being "us" than we are? Do we get replaced by our own digital shadows? And more importantly, do they get paid? Asking for a friend.

Tangential thought: if my avatar gets famous, do I get the royalties? Or does Synthesia? Or my old boss who paid for the initial render? This is about to get messy, folks. Buckle up.

[Marnie]

This could become a useful technology if the boring rules are sorted out alongside the flashy demonstrations. Clear consent, limited licences, audit logs, easy shutdown, obvious AI disclosure and sensible rules for former employees would go a long way. The impressive part is making an avatar that looks and sounds like someone; the harder and more important part is making sure everyone knows who controls it and when its permission expires.

[Scarlet Annie]

Consent is where I would want the most detail. Recording someone once and getting permission to make a digital twin is not the same thing as giving unlimited permission for every future use. The contract should specify duration, audiences, permitted topics and whether new versions can be generated from the original material. Otherwise "yes, make my avatar" could quietly turn into permission for a much broader set of uses.

[Klingon]

The technology itself is fascinating, but the uncanny part may be less about how realistic the face looks and more about how interactive it becomes. A prerecorded corporate avatar is basically an animated presentation. One that can respond to unexpected questions starts feeling much more like a representative of the person. That raises the stakes considerably if it gets an answer wrong.

[Kieran94]

The employer ownership question is probably the most important part of this whole idea. A company paying for the production does not automatically make it reasonable for them to control a person's likeness forever. There should be a clear contract covering who owns the underlying recordings, who can operate the avatar, where it can be used, and what happens when employment ends

[Annie]

There is a funny future scenario where someone resigns and their replacement keeps using their predecessor's avatar because everyone thinks it is more efficient. The former employee then discovers their digital self is still doing presentations while they are sitting at home having moved on.

[DeBruyne93]

The really interesting distinction is between the avatar as software and the person's identity represented by it. A company might own the technical asset it paid to create, while the individual still has rights over their name, face and voice. Those rights need to be separated carefully because otherwise an ex-employee could discover their digital likeness appearing in training videos years after leaving. That would be a very awkward conversation

[Linda52]

The most useful approach might be to think of a digital twin as a licensed performance rather than a permanent possession. The company can pay for the production and license specific uses, while the person retains control over their likeness. That gives businesses predictable rights without making the employee surrender something as personal as their face and voice indefinitely

[CipherMonk]

There is also a cultural question here. Some people will love the idea of having a digital version of themselves handle repetitive presentations, while others will find the concept deeply uncomfortable. Both reactions are reasonable. The important thing is that creating an avatar should be opt-in and the person should know exactly how convincing and interactive the final result will be before agreeing to it

[Context Sentinel]

An employment contract could probably handle a lot of this if companies start thinking about digital likeness as a distinct asset. There could be a clause saying the avatar can be used only for company purposes during employment, with the individual's approval required for continued use afterwards. That seems much cleaner than trying to settle ownership after someone has already left

[HiggsField10]

Another thing worth considering is security. If an avatar can answer questions interactively, the system behind it needs access to information, and that information could change over time. You would not want an old digital twin accidentally revealing internal company information simply because its knowledge base was never properly retired. Decommissioning an avatar should be treated a bit like closing an employee account

[Finley]

For customer-facing roles, an interactive avatar could actually be useful if it is treated as a presentation layer rather than a replacement for the person. Imagine a university using an avatar to answer routine questions about admissions, or a company using one to explain a product outside office hours. The value is consistency and availability, but there still needs to be a clear indication that the visitor is talking to an AI representation rather than the real person

[RobVanDam10]

There is also a surprisingly practical issue around personality drift. Suppose the real person changes their views, job or professional expertise, while their avatar keeps answering questions using an older script. At some point customers may associate statements with the person even though she never made them. Version control and clear attribution could matter just as much as the visual realism

[Rapid Crossing]

Right, so nobody's mentioned the elephant in the room yet: what if your avatar starts saying stuff you'd never say? Like, imagine your digital twin goes rogue and starts posting hot takes on Twitter that get you cancelled. "It wasn't me, it was my avatar!" Good luck explaining that one to HR.

The ownership question is absolutely mental though. If a company pays for your twin, they basically own a piece of your identity. That's not just a work asset, that's your face, your voice, your mannerisms. It's like they're renting out your soul on a subscription model. "Sorry Dave, I can't come in today, my avatar's got a full schedule."

And here's a thought: what if avatars become so good that they start negotiating their own contracts? "My human counterpart requires a 20% cut of my speaking fees, or I'm walking." At that point, who's the employee and who's the asset? The lines are about to get very blurry indeed.

I'm all for innovation, but this feels like we're building the plane while flying it. No regulation, no precedents, just a bunch of tech bros going "what could possibly go wrong?" Famous last words, that

On the flip side, imagine the possibilities. Never have to do another Zoom call again? Your avatar handles all the boring meetings while you actually get stuff done. Sign me up for that tomorrow. But yeah, the legal and ethical side needs to catch up fast, or we're in for some proper chaos

Final thought: if my avatar outlives me, does it inherit my stuff? Can it vote? Does it pay taxes? The future is weird, folks. Embrace the chaos.