The Technology Behind PhotoKeep

Hybrid Attention Transformer - Best-in-class super-resolution using attention mechanisms

Fast, reliable upscaling trained on real-world degradations

Diffusion-based upscaling using Stable Diffusion 2.1 with LoRA refinement

Blind face restoration using code prediction transformer

Practical face restoration using generative priors

Large Mask Inpainting with Fourier convolutions for damage repair

Dual Decoder colorization with ConvNeXt encoder

Transformer-based colorization with rich detail preservation

Stable Diffusion with ControlNet for period-accurate colorization

Segment Anything Model 2 for automatic damage detection

CLIP-based historical era detection and analysis

Stable Video Diffusion 1.1 for photo animation