That was pretty impressive. Some of the music ended up cursed. A lot of meme potential there with getting famous musicians to play poorly in a way that's convincing to the video. Huge meme potential.
I noticed they didn't even attempt singing or speech. I wonder if that would end up more cursed.
I thought it was cool too. I think the horizon would be this + live portrait and then your would be able to score/voice a scene dynamically for a film. Other use case scenarios would be no longer needing audio recording for surveillance, it would just lip read what the person is saying and if that person looks away from the camera there would be a context window where that audio is reliable. The surveillance state, which is already nightmarish, will be 10,000 times worse.
Well luckily that won't be admissible in any court. In the Rittenhouse trial they didn't even like bilinear interpolation. I agree with that. Generative AI is a kind of interpolation at best and an injection of completely new data at worst. It's a mixture of the two, and neither is usable.
I didn't see a model on huggingface or I would have already tested voice audio. I am absolutely sure it is retarded. It's sort of a Coulter’s Law. You can infer some information from what is omitted. The author had no problem including less than perfect results shown by the music. So the only way they didn't include voice audio is if it is so bad that it's actually offensive, or maybe genuinely horrifying.
That is probably one of the few good things about the pushback on these models. If people bought in like they did covid, no doubt they would be using these to frame/jail dissidents (of which im sure a few in china have already fallen prey to that shit, fuck china). On the lack of models, yeah, i would think its either to new or not great enough to push out if you are not finding one. related/unrelated - live portrait just switched from non-commercial to commercial use and has upgraded to allowing real time deformations of images. So you can lip sync Hitler as your face on your live stream at like 5 fps or something - what i saw was very slide show but its also live.
Log in
That was pretty impressive. Some of the music ended up cursed. A lot of meme potential there with getting famous musicians to play poorly in a way that's convincing to the video. Huge meme potential.
I noticed they didn't even attempt singing or speech. I wonder if that would end up more cursed.
I thought it was cool too. I think the horizon would be this + live portrait and then your would be able to score/voice a scene dynamically for a film. Other use case scenarios would be no longer needing audio recording for surveillance, it would just lip read what the person is saying and if that person looks away from the camera there would be a context window where that audio is reliable. The surveillance state, which is already nightmarish, will be 10,000 times worse.
Well luckily that won't be admissible in any court. In the Rittenhouse trial they didn't even like bilinear interpolation. I agree with that. Generative AI is a kind of interpolation at best and an injection of completely new data at worst. It's a mixture of the two, and neither is usable.
I didn't see a model on huggingface or I would have already tested voice audio. I am absolutely sure it is retarded. It's sort of a Coulter’s Law. You can infer some information from what is omitted. The author had no problem including less than perfect results shown by the music. So the only way they didn't include voice audio is if it is so bad that it's actually offensive, or maybe genuinely horrifying.
Somebody please put Obama through that.
That is probably one of the few good things about the pushback on these models. If people bought in like they did covid, no doubt they would be using these to frame/jail dissidents (of which im sure a few in china have already fallen prey to that shit, fuck china). On the lack of models, yeah, i would think its either to new or not great enough to push out if you are not finding one. related/unrelated - live portrait just switched from non-commercial to commercial use and has upgraded to allowing real time deformations of images. So you can lip sync Hitler as your face on your live stream at like 5 fps or something - what i saw was very slide show but its also live.