Using this method I get an “error parsing debug value” if I use motion deblur or stabilization. Works fine if I do a straight upconvert/cleaning. Definitely something I’ll be adding to my process!
Interesting tutorial, I’m going to get it on a Word file. I could use it one day.
Seems with the latest 3.3.8 update, mp4 videos encoded with h264 Dione DV are no longer streaming properly online.
this tutorial was aimed for those who want to deinterlace in TVAI and keep the fps the same. it isn’t supposed to compete or be better or less good then other deinterlacing methods. I personally deinterlace with Hybrid. but many prefer using TVAI as their sole tool. so i created this for those target audience.
I don’t have MacOS, so i don’t know how well this would work on MacOS. this was tested on windows.
on Windows or MacOS?
Yes I understood but precisely I was also talking about the deinterlacing TVAI, I did not perceive any difference in quality in 25 fps or 50 fps … weird…
I myself also did not see any quality advantage between 25fps vs. 50fps (bobbing) that is why i personally remain with x2 fps. double fps was smoother motion for me.
but this tutorial is for those who want to retain the 25fps (or whatever original fps they have) if they decide that they do see a difference. on paper, the original fps should provide higher vertical resolution, in practice, i did not see any difference.
Okay, it reassures me, in any case I also think that it is better to keep the fluidity of 50 fps, easier for models to clean the video with more images in my opinion. Okay, it reassures me, in any case I also think that it is better to keep the fluidity of 50 fps, easier for models to clean the video with more images in my opinion. But it’s amazing that the quality is the same because as you say, the vertical resolution is supposed to be limited by deinterlacing. a bit like the bitrate of a video, with an equal bitrate, a 25 fps video will be more detailed than a 50 fps video.
I compared this to chronos fast + dione tv (on old anime) and the results were unexpected. For starters the exporting time is exactly the same but the end results are not even close. The chronos one is 8.71mb while this is 9.60mb (15 seconds). The chronos one is either lagging or ahead with a single frame compared to this version. The chronos version makes it look laggy and choppy with lot of ghosting compared to no frame doubling. This tutorial wins hands down.
1 Like
No matter whether it’s 50/60 or 25/30 you have to de-interlace. So the vertical resolution won’t be any better at the higher frame rate. You just get smoother motion with the higher frame rate. If there was no de-interlacing, the 50/60 version would just be the 25/30 with each frame repeated twice (and all the interlacing still there). The extra motion information only happens because it keeps one set of lines on one frame, then the other lines on the next. Topaz seems to be doing something more sophisticated than just a basic de-interlace because the results are much smoother that what I get with other methods.
I use Topaz on source clips for edited videos and those final videos will never be 59.94fps. So for me, I’d rather not even bother getting the full frame rate if I can get 29.97 at a much better speed and file size. This method is worth a shot, but it would be great if Topaz just let me maintain the source frame (not field) rate.
I know interlacing and I know the effects it has 
I deinterlace my sources by doubling from 25 to 50, because effectively, there is no change in quality, at least with qtgmc or yadiff needi3 etc… always the same quality, and when we apply a model with vai, the more images there are, the better the results I noticed, especially since I like the fluidity, even if the video finish is twice as heavy, no big deal, the important thing is to have the quality with a good movement. afterwards free for those who want to maintain the 25/30 fps, but we can do it with topaz, it’s not complicated, I don’t really remember the procedure but Akila made a post on that, you have to ask him or let him read this thread.
Is Topaz using inter-frame information to do the scaling and de-interlacing? If so, the extra frames may be useful in the final image. If not, it should make no difference. I don’t know the answer but assumed it was doing all the processing within each individual frame per.
I don’t really know, I just think he uses bwdif
Topaz VAI is not doing any de-interlacing, it calls ffmpeg to do the de-interlacing for it using ffmpeg Bwdif. Topaz only enhance the picture/video after de-interlacing completed.
isn’t this post (1 post in this thread)?
Yes it is ffmpeg but it is bwdif which is used. The post where you explain how to deinterlace in doubling the frame rate.
if you mean in topaz, there is nothing to do/configure as this is the out of the box behavior, to double the fps.
if you mean in Hybrid, you mean this post?
I’m stupid, I didn’t see that it’s on this post where we discuss. You explain well how to deinterlace without doubling the fps
1 Like
In that case the extra frames shouldn’t make any difference. Though now I’m curious if de-interlacing it myself beforehand would give similar/same results to letting Topaz/ffmpeg do it.
If it’s just running a de-interlace from ffmpeg and then processing, why aren’t all the models available? It seems like it should be able to run them all.
in any case, deinterlacing via qtgmc remains better for me
if you going to also use bwdif with ffmpeg, you might see the same.
it’s a very good question. the same i asked why Dione AI (e.g. Dione: DV) is not available for progressive scan video
Today I compared the results of running Iris in its various interlaced/progressive modes. The result of Iris in interlaced mode on interlaced footage is nearly identical to what I get if I de-interlace the footage in advance and run Iris in progressive mode.
It’s close enough that if I toggle between the results, I can’t see any difference at all, and if I combine them using difference mode, I get a black screen. There’s probably some micro differences between exact pixel values, but not enough to ever matter.
The only real difference is the frame rate and render time. Running it on interlaced footage gives me double the frame rate and about double the render time.
I do see a difference with “Interlaced Progressive.” The footage gets a little softer with Interlaced Progressive vs the other two. It’s not a massive difference, but I can see it clearly.
That’s Iris. I did try running the other models on the de-interlaced footage. Most of them didn’t look good. Maybe that’s why they aren’t available for interlaced footage. Proteus has some potential since I can fine tune settings. In Auto it gave some unpleasant artifacts and wasn’t usable. Maybe better settings would help. It’s worth some more experimentation.
1 Like
