At the top of the Topaz Labs product page there is a short video of a person behind some leaves.
When you select Compare you can see the unprocessed video on the left, and the processed video on the right (this video has been processed by the Rhea model)
It is expected that the unprocessed video would be a lower resolution than the processed video as Rhea is a upscaling model and this is to demonstrate Rhea. But the example is misleading.
It’s misleading because the unprocessed video has a resolution of 426x240 pixels. While the upscaled result is a resolution of 3840x2160. This implies that Rhea can do a ~9x upscale and provide amazing results, when in reality it can only do a 4x upscale.
Higher than 4x upscales can be achieved by running the video through multiple upscaling passes, in which case that should be noted on the processed video if it is used.
On a side note, running the original 426x240 video through Topaz Labs with two lots of 4x Rhea results in a output video thats notably worse than the one shown on the website.
