Multiresolution synthesis
Image Analogies
Show an algorithm an image and its painted counterpart. It uses that relationship to synthesize a treatment for a different image.
Try it
Runs locally in your browser. Uploaded images stay on your device.
A and A′ must be an aligned pair. The sample A′ was computed by the painterly renderer; B is a separate photograph.
Ready when you are.
Worker processing · local inputs · downloadable output
How it works
A and A′ must be an aligned pair. The sample A′ was computed by the painterly renderer; B is a separate photograph.
- Build matched Gaussian pyramids for A, A′, and the new image B.
- Compare source appearance and already-synthesized neighborhoods to find candidate correspondences.
- Choose between appearance matches and coherent neighbor predictions, then copy pixels from A′ into B′.
Read the original research
Hertzmann, Jacobs, Oliver, Curless & Salesin · 2001
Computed examples
An aligned example pair, a new input, and the synthesized result.
Source photographs were generated for these studies. The displayed results were computed by the algorithms; they are not generated illustrations of an expected result. The painted exemplar was produced by the painterly study and stays spatially aligned with its original. Published previews are compressed for the web.
Implementation and limits
A small, bounded multiresolution synthesis study with sampled search. It does not use a neural network or infer semantic style; patch artifacts are visible, especially across unrelated scenes.
Interactive settings are bounded to keep the experiment practical in a browser. Higher detail can cost more processing time without guaranteeing a better result. The implementation and its tests are available in the source repository.