Updating an SD MiniDV project to HD in Vegas Pro 23

Dexcon wrote on 2/20/2026, 6:49 AM

There are great YouTube channels (e.g. NASS) taking film footage from the 1910s, 20s, 30s, 40s and 50s and updating that video to modern fps frame rates, stabillising the video and in many cases colorising that historic video (with some weird results where often black cars appear as purple). A common comment on those YT videos is that it's like a time machine going back into the past - I couldn't agree more.

My current project is nowhere so historic. For the past week, I've taken a nearly 19 years old project of a tour of rural Switzerland (in 2007) which I would have originally edited in ULead Video Studio. Back in 2018, I began the restoration process by matching the video events in Vegas Pro to the base Ulead Video Studio render. At the time, stabilising the 720 x 576 video worsened the limited video quality, but in 2026 that loss can be offset in VP23 by using Vegas Pro's AI Upscale FX.

In Project Properties, 1920 x 1080 was set but the pixel aspect ratio had to be reset to 1.00.

I am now able to stabilise most SD video events in Mercalli 5 SAL (several but not many events were better stabilised in VP's native Video Stabilizer).

BTW, VP's AI Upscale was set to Sharp and Cutout was set to 0.500, 0.500,1.030 for the original un-stabilised via Merc 5 MiniDV video.

Rendering is yet to happen of the 50 mins project - but at the moment on the timeline it's looking good.

And no crashes in Vegas Pro 23 b302.

Back to post

Comments

fr0sty wrote on 2/22/2026, 7:37 PM

@fr0sty Could you try these two files: https://www.swisstransfer.com/d/545cdf73-90f7-4b74-ae45-87b24e972c60

one DV and one HDV file, both need also first stabilisation, upscaling to HD is sufficient. Thanks in advance.

In order to keep memory limits in check, I had to upscale this frame-by-frame. Here was my workflow:

Export a frame sequence in VEGAS to PNG

Import PNG sequence as a batch set to increment in Comfy UI, this way each time I run Comfy, it increments the batch to the next frame.

Set the run batch counter to 100 (the max it allows) at let it rip.

From there, you can set the seed of the batch image loader to the next frame past 100 to do the next 100 frames.

Why do it this way? If I try to process the entire video at once, it will run out of VRAM even on my 32GB RTX 5090. By exporting an image sequence first, it forces Comfy to process everything one frame at a time, bypassing the memory restrictions of trying to process the entire video at the same time.

The results are OK, but they could have been much better had we properly de-interlaced the source video first. The video is progressive scan, but is very obviously interlaced, and the AI is struggling to de-interlace it first. So, if we went back to the source material for this video and properly de-interlaced it first, the results would come out much sharper. The increase in resolution is most notable on the clothes in the background.

Original:

Comfy UI Upscaled:

fr0sty wrote on 2/22/2026, 7:53 PM

The second video you posted is also interlaced, but it was not saved as a progressive scan file so I was able to de-interlace it myself in VEGAS upon exporting the frame sequence. This is how it came out:

Original:

Upscaled to 8K:

I'll share the video files once they finish processing.

Howard-Vigorita wrote on 2/22/2026, 8:06 PM

Render time on my v2 upscale/grade completed in 3 hr 33 min with vp23 b356. Project is 27 min. I did this version with the sd avi as source and tweaked color grading with a 2k upscale plus AI Upscale and AI Sharpen as event FX. Rendered with Nvenc Hevc to mp4 @ 25 mbps. YouTube then converted it to vp9: youtu.be/5HHGNukC2Jw

My own impression is that the original camera content looks much better upscaled and was well worth the effort. But the smaller fonts used in the end credits seem to be a lost cause. With more effort, I think I can now read them well enough to re-author them. I've already started replicating the rolling credits in Vegas and want to see if I can do it without any 3rd party add-ons.

Dexcon wrote on 2/23/2026, 4:22 AM

@3POINT ... have a great trip.

You know that you can apply stabilization to all selected clips on the DR timeline with ONE click? Just check afterwards the stabilization of each clip and change stabilization mode for clips which result is not as expected.

Having today done over 100 stabilisations in DR, there are unfortunately too many variables to do the stab process globally. Many shots (esp. moving shots taken from a bus or train) work better with DR's stab settings set to 'Similarity' but for most shots "Perspective' works best ... BUT ... 'Camera Lock' activated works well with reasonably static shots; with others where camera movement is involved (e.g. pan, tilt, zoom or just too shaky because the camera's [from 2003 and was Panasonic's first domestic MiniDV camera to feature 16:9] inbuilt stab wasn't very good), 'Camera Lock' is best not activated and then it's a case of adjusting the 'Smooth' slider to get the best balance between stability and the amount of zoom-in on the shot.

Incidentally, in DR I've globally set the de-interlace quality to DaVinci AI Neural Engine.

I've also compared rendered upscaling image quality between various DR AI Super Scaling options as against Vegas Pro 23's AI Upscale - and VP's AI Upscale provided a slightly better result IMO.

Overall, I am very encouraged about the results I'm getting - they're leaps and bounds above the original SD result.

Harold-Linke wrote on 2/24/2026, 2:25 AM

@Dexcon, which parameters did you use for the Upscale in VP23 with the good results?

Dexcon wrote on 2/24/2026, 3:57 AM

@Harold-Linke ... Thus far with AI Upscale, it's 'Sharp' with the 'Improve Sharpness' checkbox checked Once all the stabilisation is done in Resolve (now half done), I'll have to check if that's the best setting for all video events because I've already seen a few that 'sharp' may not be the best choice - but then there's also years' old color correction applied to those events. No doubt I'll be redoing color correction later on using VP23's CGP.

Initial testing also suggests that AI Upscale is very marginally better applied at the event level rather than at track level - using preview at Best (Full). But it's still a work-in-progress.

Harold-Linke wrote on 2/24/2026, 5:07 AM

@Dexcon, thanks. I will test it with my material.

Howard-Vigorita wrote on 2/25/2026, 12:24 PM

I should amend something I posted earlier in this thread. The source avi from my vhs tape is listed as 720x640 4:3 in MediaInfo. But if I don't check the box, "Allow source adjust the the frame rate" in the render template (2160x1440, 0.9091 aspect)... MediaInfo says the render is 3:2 which is slightly wider. If I check the box, then the preview and render line up exactly with my source and MediaInfo says it's 4:3.

fifonik wrote on 2/26/2026, 3:23 AM

@fr0sty Could you try these two files: https://www.swisstransfer.com/d/545cdf73-90f7-4b74-ae45-87b24e972c60

one DV and one HDV file, both need also first stabilisation, upscaling to HD is sufficient. Thanks in advance.

This is how my automated script processed your SD footage with default settings:

Upscale only (with AviSynth plugin QTGMC):

https://www.dropbox.com/scl/fi/psq1nbfbcthsqa81jff2c/2003-11-24_0315_u_x264_sp-abr-15000.avi?rlkey=foav0spmnt3646wkccj1qc8q7&dl=1

Upscale (with AviSynth plugin QTGMC) + Deshake (with deshaker virtualdub plugin):
https://www.dropbox.com/scl/fi/v09yvzi3du4fbhybgo93q/2003-11-24_0315_ud_x264_sp-abr-15000.avi?rlkey=bl4cqsa2lcz07ajse5ax5dsm3&dl=1

Results compressed with x264 (ABR 15000kbps).
In real life I'm compressing it using "visually loossless" Cineform codec (filmscan2 preset) and then editing in Vegas.

In deshaker plugin I'm using options that re-building cropped borders. However, this is not always works perfect, so I'm doing some cropping in Vegas later (you can see these artifacts on the right in resulting video).