Masking specific partial areas for less Enhancement, and-, or masking areas for more Enhancement would be a killer super nice new feature. often there are areas where recovering detail slider in enhancements screws up a face of a person beyond recognition, but the amount of recovery details is indeed very much needed in that strength for the rest of the image.
masking certain areas for different amounts of enhancement would be an amazing step forward.
masking an area, for example again a face, which already had turned into an unrecognizable thing, and then being able to return just that area of that face, to a previous version, with a mixing slider for example, again, just for that face area that is masked, would be even a better solution.
but that is probably way to difficult to code and program, right?
Use an NLE to do this to your output product, if quality > time matters more
Eventually we’d want the models to learn this stuff anyway
For instance, I had to drop in a whole 10-second scene using a different model (Proteus instead of Iris) because of the way Iris thinks a deliberately defocussed person is a problem it has to solve, sometimes. It just needs to be cleverer and leave it the hell alone, and then I’ll just use Iris.
Other stuff … I don’t expect Topaz to ever learn. Maybe it’s too esoteric, or too specific to one project. The right thing to do will always be touch-up and editing. Automation increases what we can do, instead of reducing the work we have, right?
I think this idea is a good one, provided we get some other ideas in place too.
It would be awesome if TVAI produced artifact-free results every time, but that’s more unlikely the longer the video.
The ideal way to use TVAI would be to have multiple sections. You could define where each section starts and ends. What models will be ran—and define masks to ignore areas of the video—all per section.
If we can animate those masks, that would be the ultimate tool.
At the end of the day, TVAI will be just another tool, another filter (with lots of fine-tuning settings).
If you’re adding it to a toolkit that you already use a lot of other tools in, it’ll fit fine.
AI probably feels a bit “special” this decade, and can really soak up your CPU/GPU unusually, but in the scheme of things, the heavy resources needed for video, HD, 4K, HDR, it’s going to fit right in.
Which is just to say that you’ll be manually masking, disabling/enabling, etc, forever into the future.
If not for some instances, then definitely for others. The more professional your work, the more likely.
