Multitalk: Generate high-quality portraits using pretrained weights

Step-by-step

1. Open the model instantiation code within the DeepInverse framework
2. Modify the lines of code responsible for loading weights to point to the FFHQ dataset
3. Run the sampling code to begin the diffusion process
4. Wait for the final iterations of the diffusion to resolve the noise into clear images

Options

• Swap in pretrained weights from different datasets by changing one or two lines of code

Watch out for

• The first half of the diffusion process will only display noise and will not show recognizable features

Highlights

“these faces are almost indistinguishable from real pictures”