The new vectorscope is awesome! I agree that the blue markers are difficult to see. Just wondering, would RGB values in the tooltips be useful to anyone else? I’m wondering about comparing colors quickly between GIMP and Shotcut without having to export a frame.
Preview scaling is phenomenal. I don’t have any bugs to report, but I have results from two tests that illustrate the performance gains:
Test 1: Fading
Setup:
- Video clip on V2 overlaps video on V1 by four seconds and fades in the whole four seconds. Hardware is 16 threads @ 2.4 GHz.
Results:
- 4K timeline, 4K sources, no scaling: cannot play real-time
- 4K timeline, 4K sources, 360p scaling: single-clip playback is easily real-time, but fade section stutters badly
- 4K timeline, 360p proxies, 360p scaling: the entire sequence never tops 5% CPU! Incredible!
Test 2: Track Stacking
Setup:
- Put any video clip on V1. Same video clip goes on V2 overlapping the clip on V1, but has audio detached and deleted, a color grading filter added, then an opacity filter at 20% added. This filtered clip on V2 is then copied-and-pasted to V3, V4, V5 as high as possible until the audio from the clip on V1 starts to stutter.
Results:
- 4K timeline, 4K source, no scaling: stutters with V1 alone; can’t stack
- 4K timeline, 360p proxy, 360p scaling: stacks to 21 tracks!
If I replace the color grading filter with a Text: Simple filter, it stacks to 15 tracks.
The opacity filter is critical to the test because I believe Shotcut still has an optimization that says “if V5 is fully opaque, then don’t bother compositing V1-V4 since they can’t be seen anyway”. The opacity filter forces compositing and activation of all filters across all tracks for a more accurate measure of performance.
For previous projects, I had set the video mode to 360p using 360p proxies to get this kind of editing performance, and I see improvements in the new Preview Scaling feature that beat the old method. I was surprised to see Preview Scaling maxing out 8 of the 16 cores, whereas the old method would usually use 4, maybe 6 on a good day, but none at 100%. Also, the old method was only able to stack 18 tracks whereas Preview Scaling can do 21 (this is using the same hardware for both tests), so this is win-win all the way around. I’m very excited, as you can tell.
Specific note to @Earlybite (I hope I’m not breaking the thread rules by commenting on this since the stuttering thread was closed): You asked about performance of a 4-thread processor vs. a 12-thread processor. These test results suggest that a 4-thread processor with proxy sources could possibly be sufficient in 360p scaling mode. However, with 12 threads, proxy generation and export encoding will go significantly faster, and the preview while editing at 360p scaling should also benefit significantly since the preview appears able to utilize 8 cores efficiently. Let’s open a new topic if you want to discuss further to avoid cluttering this beta conversation.