Image inpainting quality can be dramatically improved by inverting the input image into its corresponding noise representation and blending it with mask-region noise, so the background semantically influences foreground generation and eliminates boundary artifacts.

50:42mean by like this noise in the background the rest of the image. So in den noising we started with uh when during training with a noisy image and then we progressively removed noise and end up with a clean image. In image generation we also do the same thing right we start with a noise and then end up in with a clean image. But think about other process like the

51:11way that we actually train those models. Those models when we train we start with the clean image and then add noise noise and at the end it becomes like noisy image. So think about the reverse pro process where we actually train the model. Um so we can take this input image and map into noise. So we are progressively inverting a real let's say clean image into noisy versions. So kind of this

51:43thing you know very well studied understood and it's very fast like 60 mcond we can take a large image and then end up uh create going through the reverse uh den noising and a noisy version of it. So this noise is not random noise anymore. It is specific to the input image. So if I change the input image the noise is going to be different. So this is about

52:13entire image and then we have this mask of the bird. Now I added noise to there for the bird because I allow I want to allow algorithm to generate a new bird compliant with my uh field prompt text prompt. So now I changed the way that we kind of create this image. uh but I don't need to change the original model. I can still you know I change the way that I

52:40initialize the diffusion in this uh painting. And when we do that when we start with you know like this inverted noise plus the you know mask and new noise within the mask. Now uh first of all uh we can retain background but we also allow background to slightly uh uh impact the foreground like the mask itself. allows seamless harmonization uh and it generates you know high quality images but most importantly

53:16there's no boundary artifacts anymore because we have two noise you know like no like random noise and inverted noise within the same uh when we started the image we are not starting from like just noise within the mask. Awesome. Awesome. So, uh, as we suggested starting up, um, you know, Qualcomm always has a ton of papers at CVPR. We can't cover all of them. We've covered a handful of the

53:44most important image generation papers. Uh, but there were also papers on video generation. There were a ton of demos. Anything in particular you'd want to uh, call out in terms of, you know, other things folks should uh, should look for? >> Absolutely. I'm very proud of the three video generation papers because it makes video generation uh accessible to everyone. You can run such models now on your PC, laptop or