Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: Infinity – Realistic AI characters that can speak

news.ycombinator.com

Post reply on HN
Print view View original

101–110 of 320 posts

Re: Show HN: Infinity – Realistic AI characters that can speak

#101

Post by bee_rider »

jl6 wrote: post #87

Say I’m a politician who gets caught on camera doing or saying something shady. Will your service do anything to prevent me from claiming the incriminating video was just faked using your technology? Maybe logging perceptual hashes of every output could prove that a video didn’t come from you?

These sort of models are probably going to end up released as publicly available weights at some point, right? Or, if it can be trained for $500k today, how much will it cost in a couple years? IMO we can’t stuff this genie back in the bottle, for better or worse. A video won’t be solid evidence of much within our lifetimes.

Re: Show HN: Infinity – Realistic AI characters that can speak

#102

Post by lcolucci »

klipt wrote: post #93
marginalia_nu wrote: post #76

Tried my hardest to push this into the uncanny valley. I did, but it was pretty hard. Seems robust. https://6ammc3n5zzf5ljnz.public.blob.vercel-storage.com/inf2...

It even works on animals: https://6ammc3n5zzf5ljnz.public.blob.vercel-storage.com/inf2...

I think you've made the 1st ever talking dog with our model! I didn't know it could do that

Re: Show HN: Infinity – Realistic AI characters that can speak

#104

Post by lcolucci »

archon1410 wrote: post #99

The website is pretty lightweight and easy-to-use. The service also holds up pretty well, specially if the source image is high-enough resolution. The tendency to "break" at the last frame happens with low resolution images it seems. My generation: https://6ammc3n5zzf5ljnz.public.blob.vercel-storage.com/inf2...

Thank you! It's interesting you've noticed the last frame breakdown happening more with low-res images. This is a good hypothesis that we should look into. We've been trying to debug that issue

Re: Show HN: Infinity – Realistic AI characters that can speak

#105

Post by sidneyprimas »

RobinL wrote: post #56

Have to say, whilst this tech has some creepy aspects, just playing about with this my family have had a whole sequence of laughs out loud moments - thank you!

This makes me so happy. Thanks for reporting back! Goal is to reduce creepiness over time.

Re: Show HN: Infinity – Realistic AI characters that can speak

#107

Post by drhodes »

andrew-w wrote: post #94
drhodes wrote: post #92

Earlier quoted context omitted.

just an idea, but what if the appended audio clip was reversed to ensure continuity in the waveform? That is, if > <PILC.

This is exactly what we do today! It seems to work better the more you extend it, but extending it too much introduces other side effects (e.g. the avatar will start to open its mouth, as if it were preparing to talk).

Hmm, maybe adding white noise would work. -- OK, that's quite enough unsolicited suggestions from me up in the peanut gallery. Nice job on the website, it's impressive, thank you for not requiring a sign up.

Re: Show HN: Infinity – Realistic AI characters that can speak

#108

Post by ilaksh »

sidneyprimas wrote: post #98
ilaksh wrote: post #96

It would be amazing to be able to drive this with an API.

We are considering it. Do you have anything specific you want to use it for?

Basically as a more engaging alternative to Eleven Labs or other TTS.

I am working on my latest agent (and character) framework and I just started adding TTS (currently with the TTS library and xtts_v2 which I think is maybe also called Style TTS.) By the way, any idea what the license situation is with that?

Since it's driven by audio, I guess it would come after the TTS.

Re: Show HN: Infinity – Realistic AI characters that can speak

#110

Post by ilaksh »

andrew-w wrote: post #49
dorianmariefr wrote: post #34

quite slow btw

Yeah, it's about 5x slower than realtime with the current configuration. The good news is that diffusion models and transformers are constantly benefitting from new acceleration techniques. This was a big reason we wanted to take a bet on those architectures. Edit: If we generate videos at a lower resolution and with a fewer number of diffusion steps compared to what's used in the public configuration, we are able to…

Wowww.. can you buy more hardware and make a realtime websocket API?
Post reply on HN