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

Not robust enough to work against a sketch https://6ammc3n5zzf5ljnz.public.blob.vercel-storage.com/inf2...

though perhaps it rebelled against the message

I had difficulty getting my lemming to speak. After selecting several alternatives, I tried one with a more defined, open mouth, which required multiple attempts but mostly worked. Additional iterations on the same image can produce different results.

Cartoons are definitely a limitation of the current model.

Nice! Earlier checkpoints of our model would "gender swap" when you had a female face and male voice (or vice versa). It's more robust to that now, which is good, but we still need to improve the identity preservation

The jaw is particularly unsettling somehow.