He mentioned wanting to include pictures that would take advantage of full 1080p resolution. You see 2 methods, where SD type resolutions are included. 1 is using borders, maybe equivalent to 720p, and the other is full upscale to 1080p but you have an obviously soft image, but if primarily viewed on small screens maybe appropriate. You could use mild sharpening if original wasn't sharped/over sharpened to begin with. Sharpening generally makes people's faces look ugly. It can have it's uses for bringing out detail in certain situations.

You have to set properties to 1080p or your pictures won't be 1080p, and need to change aspect ratio of your dvd footage to match 1:1 of 1080p. There might be a specific filter for that, or it could be done with event pan crop, where you unlock aspect ratio and correct for the DVD AR. I have no actual experience with dvd, so not speaking from experience. Many here work with dvd even today

If you "must" upscale DV, go no larger than 1280x720. Use Lanczos scaling in another app if possible..

If you use .9091 PAR, it will give you unaltered 4:3.

I "have"widened it a bit to go better with full frame stills by using 1.0 PAR, without making it ^too" fat.

If it's not going to YouTube, make your video natural resolution, and let hardware upscaling do the work, which is still preferable.

Don't expect great results. SD tape is what it is.

@Carl-G

Here's an example of what I've done in the past mixing SD with photos.

The key to making the video look acceptable (IMHO) is high quality bob de-interlacing. In other words, upscaling your 480i video (I presume your video is interlaced.) to 720p. You will need to venture outside of Vegas to achieve this and use something like the QTGMC plug-in inside of Avisynth. I should add that QTGMC de-interlacing is supported in HappyOtterScripts, although the included scripts would require modification to achieve the best results for your application. Let me know if you have any interest. HOS is still in beta so you can download a trial.

If this is just for computer viewing (the ones you listed above) why bother upscaling it to 1080p. Render it as a 873 x 480 resolution with aspect set as 1:1. There is nothing to be gained by uprezzing to 1080 for the smaller screens. Let the computers upscale it as they play it. It will retain better quality.

 

Makes sense. Why upscale to only downscale... they did want one presentation version on a large screen.... I'll go with AVISynth for conservative treatment, then redo the mix on the timeline.

There's also these neural networks that actually do add artificial detail to the image, and sometimes can do a quite amazing job of doing it.

https://captrobau.blogspot.com/2019/05/tutorial-upscaling-video-with-topaz-ai.html

@wwaag, I see this uses ffmpeg, is this or can this be integrated with HOS?

Garbage in - garbage out :)

However, you might be interesting to check this (avisynth, I'm using the technique).

So far the AI upscaling is actually starting to challenge that GiGo philosophy. Here's another example from an old video game they completely remastered using AI upscaling:

PAL DV footage upscaled using AI. Blows the traditional methods out of the water, and definitely higher quality than just letting the computer do the upscaling by leaving the video native res:

@Carl-G

Here's an example of what I've done in the past mixing SD with photos.

The key to making the video look acceptable (IMHO) is high quality bob de-interlacing. In other words, upscaling your 480i video (I presume your video is interlaced.) to 720p. You will need to venture outside of Vegas to achieve this and use something like the QTGMC plug-in inside of Avisynth. I should add that QTGMC de-interlacing is supported in HappyOtterScripts, although the included scripts would require modification to achieve the best results for your application. Let me know if you have any interest. HOS is still in beta so you can download a trial.

Very Nice! Thanks for the reminder about Bob De-Interlacing. I've used that before with great results. I will look into the QTGMC - sounds interesting.

So far the AI upscaling is actually starting to challenge that GiGo philosophy. Here's another example from an old video game they completely remastered using AI upscaling:

PAL DV footage upscaled using AI. Blows the traditional methods out of the water, and definitely higher quality than just letting the computer do the upscaling by leaving the video native res:

This is amazing! I'm a big fan/user of Topaz in Photoshop, but I was unaware of video applications. Your example is great! I will look into this also! I can see a big use for this in future projects.

I'm actually looking into it for some of mine as well, as we have some compilation projects that are commemorating a decade or more worth of festivals we've filmed, and those older festivals are 720p... going to uprez to 4K using AI and see how it looks.

Anyone tried the Gigapixel AI that was mentioned earlier?

I just played a little with my footage I used for testing SimpleSlugUpscale:

- Gigapixel AI is image upscaler, so you will need to extract images first (can be done in command line with ffmpeg)

- For interlaced source you will need to deinterlace it first. Good deinterlacer is the crucial part of the process. With SimpleSlugUpscale, QTGMC is already there. I tried yadif & bwdif that are in ffmpeg.

- It is extremely slow (not sure if this is in my case only or at all)

So here you are (comparison frame 100, you would better download frames and compare them in your image viewer):

Original frame after yadif:

Upscaled with Gigapixel AI (default auto options):

Upscaled with SimpleSlugUpscale (UPDATED) (settings: drate=true, outwidth=1024, DARin=4.0/3.0, DARout=4.0/3.0):

Previously added SimpleSlugUpscale image (with slightly incorrect aspect ratio caused by missing DARin):

Sorry SkyNet, but I'm not impressed. Do not like what you did with water and leaves (some kind of oil painting). And why you broke my paddle?

Commands I used for extracting frames:

ffmpeg.exe -i Source.mpg -vf "yadif=1,scale=768:576:flags=lanczos" Source\Source%04d.png -hide_banner

ffmpeg.exe -i Upscaled.mp4 Upscaled\Upscaled%04d.png -hide_banner

The complexity of the scene may be throwing it off, that's definitely a good stress test for it. Are you sure you uploaded the original frame? It looks like the AI upscale again.

The cool thing about these AI upscalers is they will constantly improve as the neural networks are trained.

I uploaded correct frame (frame size 768×576, while others are 1024×768). It is NOT from the source directly (source is interlaced). It is after the first command extracted fields and deinterlaced them.

> The complexity of the scene may be throwing it off

Well, this means it is unusable :)

As with any machine learning, if the system is trained on some samples it do good job (like with the game mentioned).

If it is not trained -- it cannot do anything at all. Unfortunately, you cannot train it on all different kind of waves and leaves. It is not possible.

I was interested in understanding the best way to handle this in Vegas as well.

Here's my summary of what I got so far.

Q: What's best way to tackle this in Vegas? A, B, or C?

A: Don't attempt doing this in Vegas try A, B, or C instead outside of Vegas or use a script inside of Vegas which allows you to use A, B, or C outside of Vegas.

Really starting to rethink my decision that since I was familiar with working with audio in Vegas that Vegas would be a no brainer for me to start working on some video projects but so far a lot of the advise I see on these forums seem to recommend doing things outside of Vegas, making me wonder what you actually do inside of Vegas?

For me, there are two things I'm doing outside of Vegas: upscaling and encoding. I have not done the upscaling for a few years already as I do not use SD camcorder any longer and almost all old footage already processed. You do not have to do the upscale outside of Vegas. It is just a way to get a bit better quality.

Everything else I'm doing in Vegas (sure I'm using 3rd party plugins).