This machine learning model is for the Nuke’s copycat/inference tool. It attempts to convert common images of limited dynamic and bit-depth (x3 rgb 8bit). Into a image with more range that float/exr CG application or high dynamic range camera might produce.
It specifically recovers:
- clamped highlights (maintain highlight detail when graded by colourist)
- 8bit banding of colour (pass common broadcast quality checks )
Artists might source images of low fidelity jpegs/pngs etc. from:
- web images i.e. from Flickr
- generative AI image software
This model cannot guess accurate detail (as how an AI in-fill model does). It will however provide enough information to pass most stringent QC checks.
Instructions:
- Download the model / .cat file from github here.
- In Nuke load the model file with the inference node
- Place the inference node either side of an OCIOColorspace node.
The inference node working color-space must be :
<< “Input – ARRI – V3 Log3C (EI800)”
<< before being returned to the scripts working space (normally ACEScg)
Methodology
It’s a fairly simple cat.. Trained in Nuke on freely avaliable Arri Alexa sample footage from Arri’s website. It was trained in 2kDCP arri log with the copycat main input being Posterized in ACES sRGB – Texture, and the ground truth being the original image
CopyCat settings:
- Model size: large
- Crop: 512
- Total steps: 50000 (aprox day of training on RTX3090)
Caveat
For commercial jobs remember to check all images used have an unrestricted licence attached to them.
Leave a Reply