So I restore old music videos using Topaz and I’ve done over 600 in just over a year. I’ve learned lots during that time but I’ve settled down to a process that encompasses all the variations I’ve encountered. My default output is 1080p & 60 FPS and I’m using 576p sources.
My default models are Proteus v4 and Apollo v8. Chronos Fast v3 mucks up song titles sometimes, whereas Apollo never does. Apollo should be the default.
I’m here to show you two scenes where Proteus v2 is the model I’ll be using.
When there are supposed to just be lines, Proteus 3 and 4 and all of the Iris models gives us Arabic looking text… I literally screamed out " Hallelujah!" when I discovered Proteus 2 doesn’t stuff up the lines.
You’ve done goofed with the models I think. Can’t you go back to where you were with Proteus 2 and then train the new model to do the lines correctly?
So, for videos that have the line problem, I create 2 outputs to use in Premiere Pro, one using Proteus 2 and the other Proteus 4 and I’ll use the best model for the scene. This doubles the files I need to store and I’d much prefer to just use the one model!
Is there some way to combine the best of Proteus 2 and Proteus 4? Proteus 2 has the lines ability that v3 and v4 just can’t handle…
1 Like
Nothing to do with an answer, but how do you select your model ?
Did you make a table with all releases ?
I would like to try out, because I have the same problem with Proteus.
Well, actually, yes! Today is your lucky day, here’s a list of 520 videos with the AI model I used as one of the columns… there’s links to the Youtube video result, as well as the source if it is publicly available:
1 Like
I should add that I prep all my videos before I give it to Topaz … I denoise all my stuff with Hybrid (before that it was Handbrake) and get it as perfect as I can before giving it to Topaz. All Topaz has to do is to the upscale on auto with Recover detail on 100 and I’m usually happy enough with the result. Any tinkering can happen in Premiere Pro.
Godlike, I didn’t even think you build such table fo all your encoding.
I think I expressed myself badly, in fact I wanted to know which version of Topaz Video AI I had to download to get the models v1 v2 v3 aso …
I would like to do tests and comparisons like you, because I come across exactly the same problems, both the story of the teeth and the straight lines transformed into hieroglyphs.
You need to activate the previous models option. it’s in preferences, “Enable previous model versions”
1 Like
