IntrinsicAnything: Learning Diffusion Priors for Inverse Rendering Under Unknown Illumination

Generate intrinsic images (Albedo, Specular Shading) from a single image.

Tips
  • You can check the "Auto Mask" box if the input image requires a foreground mask. Or supply your mask with RGBA input.
  • You can optionally generate a high-resolution sample if the input image is of high resolution. We split the original image into Vertical Splits by Horizontal Splits patches with some Overlaps in between. Due to computation constraints for the online demo, we recommend Vertical Splits x Horizontal Splits to be no more than 6 and to set 2 for Overlaps. The denoising steps should at least be set to 80 for high resolution samples.
Type

Automatic foreground segmentation
5 200
1 10
Examples (Click one to start!)