Finally, an update again that actually brings real improvements.
Grain has been sensibly reworked and is now finally truly usable. As far as I can tell so far, the result looks much more natural. It is genuinely an improvement over all previous versions.
However, I am missing explanations within the program itself for the different options of the Grain filter. It would be nice if these were added in the near future.
I still cannot fully understand the Engine Refinements after several tests.
With the Starlight Mini model, I honestly cannot detect any change in speed. If there is any difference at all, it is only within a very small percentage range.
The image quality of the model also remains unchanged compared to previous versions.
With the Starlight Sharp model, however, the situation is different—at least in terms of speed. The model has become significantly faster, although I cannot say exactly by how much, as it strongly depends on the resolution of the source material.
As an example:
With my 1-minute test clip at a resolution of 320 × 240 and 15 frames per second, a 4× upscale to 1280 × 960 now takes only about 13 minutes instead of roughly 20 minutes.
Upscaling the same test clip by a factor of 3 to 960 × 720 now takes only around 6 minutes, whereas it previously took a little over 10 minutes.
In terms of image quality, however, very little to nothing has changed. Due to the increased speed at factors x3 and x4, I now increasingly prefer this model over Starlight Mini—also because the result does not have that “wax filter” look.
I could not detect any improvement in speed or image quality for the other models. Proteus, Iris, and Artemis remain unchanged.
Overall, this is an update that is finally worth it again after a long time and does not represent a step backward. Unfortunately, it is rather sad that such an update is now perceived as something positive. For a subscription-based model, one simply had to wait far too long for this.
