Some cuts are becoming blurred / blended between two different shots
There should be scene detection to stop this from happening
1 Like
+1 to this. I have a few videos that have some missing frames and it would be awesome to be able to use a “Repair” model for specific sections of the video.
Though I wonder if using slow motion in that clip would generate enough in between frames and then those frames could be inserted into the original video? I guess that would be a lot of work.
1 Like
Slomo doesn’t work with dups and dropped frames as it results in random speed changes, which can be more disconcerting than the original problem.
You can convert to image sequences and manually identify which frames were dropped, delete any unwanted dups, then use TVAI to generate the missing frames (I needed to load 3 frames each time for that to work), and finally, renumber the lot… I’ve done it and it works. But it takes a horrendous amount of time to fix more than a few dodgy frame sequences like that.
I think the scripts I linked to are intended to automate that manual process but I haven’t had chance to try yet, and I suspect that it will still be a slow process.
1 Like
Why don’t you use Flowframes with De-Duplication enable ? 
It is a free AI Video Interpolation software with built-in Frame De-Duplication, speed compensation and scene change detection feature. It can detect & remove duplicated frame automatically and regenerate those frame with AI interpolation. You can also adjust the sensitivity for Frame De-Duplication.
3 Likes
CG Output is still non-existent.
The Giaia → CG does almost nothing, maybe even nothing.
Still have use to version 2.6.4 when working with CG.
The latest ones. 8th of December.
1 Like
Thanks!
I, too, used to use FlowFrames for a finished project in my pipeline, but since using the de-M-Blur and apollo model, I no longer need to use it for interpolation!
I have the Patreon version, BTW…
1 Like
I use DupeGuru to nix those pesky dupes!
Please make it possible to see input frame number in “Trim” menu.
I don’t know how I can continue processing if output images are in frame numbers…
2 Likes
Can you explain a bit more? Which 3 frames are you loading into TVAI?
For example if I have 24 frames, and there are a few missing frames within them, are you saying that I’d place 3 frames before the missing frame and 3 frames after the missing frame? Are those 6 frames are enough to generate the missing frames in between? I’m also not sure how to know how many frames are missing. I guess I’d have to test out different slowmo speeds to generate what I need.
I also want to say that even without having scene detection, I still prefer the results of Apollo over FlowFrames.
1 Like
This. So THIS.
Frame number input is a must-have.
@paul-4464
About the de-interlance of Resolve.
It’s so good that all I can see is the compression of the video, it seems like it’s completely removed.
And with Proteus you can fix the compression afterwards and upscale.
3 Likes
I think what I did (a few weeks ago) was - from a copy of the video clip I was fixing - make a list of the missing frame numbers (identified by frame by frame advance, and observing any sudden change in movement e.g. a slight jump) because there was no regular pattern.
Then, separately for each missing frame, I loaded into VEAI the frame before and after the ‘jump’ and tried a 200% slowmo but that didn’t work , but it did generate the ‘missing’ frame when I loaded the frame before and the 2 frames after.
That left more frames than just the one extra that I wanted, so I had to delete the 2 or 3 unwanted ones from that litle slomo clip I’d just exported, before dropping the newly generated one into a second copy of the original clip. You have to keep an eye on the frame numbering to avoid getting them out of order, so renumber as necessary to leave gaps for each extra frame you’ll be creating e.g. 01 03 05 07 09 11, then number the extra frame to say 04, drop it in, and finally when you’ve done them all, renumber the whole lot sequentially.
It’s a pain and very time consuming. So we need that script if it works, or TVAI to be able to handle it - when they’ve debugged it! @lhkjacky I’ll try flowframes to see if that’s the answer - it needs to detect and fix irregular sequences of dropped and duplicate frames, including 100% correctly static sequences (and not remove those as they aren’t dups) - so can it, that is the question.
1 Like
Hello . . .
performance:
on my i7 8core and gtx3070 running both on 30%, → 10frames/sec … so it may have chances for improvements. my videos are DVD quality aprox 720x430 in a 2GB h264 file. i double the resolution with proteus.
“videoai=Enhanced using prob-3 with recover details at 27, dehalo at 23.0, reduce noise at 1.0, sharpen at 25, revert compression at 44.0, and anti-alias/deblur at 8.0. Changed resolution to 1432x872” “***.mp4” (section information from ffmpeg metatag )
hints:
so you running a very simple ffmpeg command, i dont copy the whole lines but i miss some useful options.
- strong-intra-smoothing=0
- aq-mode=3
- ref=5
- bframes=5
- g 60 (frames)
- rc vbr_hq
- mbd rd
- psy-rd=1 and psy-rdoq=2
- some kinde of control -cq 20 -qmin XX -qmax XX (same option exist with minrate, maxrate)
- may an option to resize with cuda ? “scale_cuda”, but its limited filter-options
- running more than one graphic-card, so i have 2 different (fast/slow) so if i running both its only the double FPS from the slow → may an option to calculate more frames on the fast one or calculate the HVEC on the slow one …
the affects i can only estimate 
but may some ideas
THY
Always good to have friends. 
This is purely speculation on my part, based on experience with other forms of digital graphics.
This may be an artifact of having a video source that is oversaturated, or out of gamut Or, the source video level and gamma are not adjusted to industry-standard video levels. (I have also had this problem when attempting to enhance old 1950’s vintage Technicolor movies. Like the Kodachrome images from that era, the color images appeared to be much more robust than the original photographed objects. )
There are two possible ways this can be fixed.
The first is to run the source through a video editor/processor that will adjust the video levels to fit the current “standard” HD
range.
The 2nd method is to run the source through a video editor/processor with both the contrast and the saturation reduced. - If using this last method, you can put the saturation and contrast back in after your rescaling and enhancement is done.
I strongly recommend trying the 2nd method.
This version doesn’t even work for me anymore. Unusable. It always says:
The drive is out of space. Really? I got one hard disk where there is 12 TB (Terra Byte) free.
Come on.
Cannot recommend it anymore to anyone.
I suspect you haven’t changed all the paths in the Directories tab in the Preferences menu to folders in that 12TB drive. If you have, then something is truly broken.
