When triming a video to export it. the video hangs at the begining. Just like the previews did before you removed audio for the previews. Triming is esencial for using the software because a lot of the time you don’t need 3 hours of video only a few minutes, and then you have to take it into an editor to re-trim the video. Other than that Video AI is improving quite a lot and getting faster to use. I can’t wait for new models that improve the monster face proplem. Maybe make the faces a little focused without the full on effect of the models. Like maybe face detection options with adjustments to reduce noise and sharpen to improve quality without the AI models smoothing them out and distorting the face completely.
Still, preview for some 23.976 fps content is out of synch. This doesnt happen for example with Alpha v3.2.1.3a (that’s just a version that I still have in parallel installed for testing).
The copy button on the Show Preview/Export Command dialog doesn’t seem to work. I have to select the text with the mouse and press Ctrl+C to copy the text to the clipboard.
Either the benchmark changed or performance tanked for 2x and especially 4X Artemis + Proteus and also on APFast on my AMD 7950X + 4090 system (I checked with parallel installed recent alpha and it performed normal, so it’s not a system problem).
Update: just re-installed 3.2.5 without reboot and performance in 3.2.5 is fine
Little question: Will you also include one day the compatibility of HDR10 files? because I just tested on a movie in HEVC HDR10 and the preview image is as bland as if I read the movie on a PC without HDR, HDR10 or HDR10+ playback compatibility.
@topaz257@david.123
For users reporting benchmark differences.
Color correction is now part of the benchmarks now. This makes the benchmark results for enhancement models is closer to the actual processing.
Color correction is largely CPU bound and applied on the output, so the performance for 2x and 4x reflects the impact of it more than the 1x on higher end GPUs.
Originally, the benchmarks were designed only to show model performance without encoder/decoder and post processing. This led to differences in benchmark fps and observed fps on machines with slower CPUs or powerful GPUs. This change to benchmark is more inline with the processing times seen by the users.
Thanks for the explanation, which suggests no change to actual processing speeds between 3.2.5 andx 3.2.6, if I read your comments correctly.
However, the changes do now make it hard or impossible to compare new and old results in the 3.2.x benchmark thread, something that users who, like me, are considering system upgrades find to be a valuable resource.
I suggest starting a new benchmark thread to enable future comparisons to be done on a level playing field.
Why a same/existing AI Model (already downloaded and exported successfully) downloading each and every-time in a batch Export… This is very annoying… Pls help… Thanks