That is a pretty big change in file size. Any reason why? I’m on Mac OS.
1 Like
the app was rewritten from scratch. the old software enhaced frame by frame as if the video where a collection of million of individuals images. the new version threats the video as a whole enhacing the video more constantly and better. it like compare windows 98 to windows 8.1. (same thing but totally new software)
1 Like
Where did you get that information?
1 Like
was informed by the devs themselves to all the software owners + i am a beta tester user and we where told to debug this new software version. i can look the forum thread where this was explained if you need it.
Nah, I had never heard anyone say anything like that before and would have never guessed it. It still just generates commands and sends them to the TVAI enabled ffmpeg. My guess would have been it just ships with more models included by default.
2 Likes
as we where explained, the old AI took frame by frame enhaced it, making it slower and less realiable the results. this takes all the frames and gives a shared or similar enhacement making the result better. it is based on old 3.5.4 software. but its like compare gigapixel to photo AI where it does the same in a similar manner but its a brand new software all thogheter.
I remember that statement about VEAI 2.6.4 and the change to TVAI 3. That’s how the Apollo model became possible.
yes but the software its not just the models. the models can be entirely different from app version to version, thats why in every upgrade the old models are deleted and replaced by new versions. the new video AI works internally in a different manner than V3 and even more than V2. how the video is processed, etc. in fact the version 3.9.1 beta had a bug where the AI lost its mind processing some videos and the output would be a fever dream of all colors/rainbow shaped images and was funny to put to the test, even that in some cases the colors where so rapidly changing that you would get dizzy on a high FPS video.
The TVAI 4.0.4 when cleanly installed directly from the DMG weighs in about 370MB. Not much different to 3.5.4
Only after you did encodes with different models and those model files are downloaded the file size grows to several GB - and this as well for both 3.5.4 and 4.0.x
1 Like
yes. the models are quite heavy it can be a few gbs in size. but 16 gbs seems to be a bit more than normal. maybe temp files/a failed encode?
At least regarding file size: Yes it is (nearly only the models). Also, there never was a 3.9.x?
there was a beta version long long ago, while they internally developed V4 but i remember there was a 3.9.5 version (not sure it if was beta or stable version)
I was a beta tester at that time and no, nothing higher than 3.5.x for the 3-series.
And at that time we already had a very early 4.0.0 Alpha version.
My TVAI 4.0.4 install currently consumes 7.2 GB with 6.8GB being model files and CoreML cache. And I’ve not even done Nyx 2 and some other models (Apollo, Chronos), so if you apply them in different resolutions and with different RAM settings I do believe that you can reach 16GB.
with all due respect but i am an older user and for a short time they developed the 3.9.x version untill the decition was made to do the V4 witch is mostly based on 3.9.x where they started to redo the software or reshaspe/rethink it. yes it was just beta releases. then they abandoned 3.9.x for a while then the V4 was started. if you need different versions for different resolutions of the same model then yes many gbs can be burned out in copies of the model in different resolutions.
