Topaz Video AI: Scaling issue with 4:3 SD (720x576) to 4K (2880x2160) // Resolve Roundtrip

Topaz Video Bugs and Issues
1

Hi everyone,

I need some technical advice regarding an issue I’m encountering with pixel aspect ratios (PAR) while upscaling in Topaz Video AI.

My Workflow:

I am doing a roundtrip between DaVinci Resolve and Topaz Video AI. I export the clips from Resolve, process them in Topaz, and then bring the Topaz output back into my Resolve timeline to match it against my reference.

My Goal:

I am trying to upscale source material from 720 x 576 (classic PAL 4:3) to its equivalent 4K resolution in a 4:3 aspect ratio, which is 2880 x 2160.

My Settings in Topaz:

Output Resolution: Since there is no built-in 4K preset for 4:3, I am using Custom Resolution and manually entering Width: 2880 and Height: 2160.

The Issue Group (“Pixel Type”): I am experimenting with the options between Original Pixel Type and Square Pixels.

The Problem in Detail:

1. Using “Original Pixel Type”: The final output ends up being way too large/distorted and doesn’t match the intended framing in Resolve at all.

2. Using “Square Pixels”: This gets me much closer. The height (2160) matches perfectly, but the width is still slightly off. I currently have to slightly downscale the width in Resolve to make it align perfectly with my reference layout.

3. Deinterlacing Only (No Scaling): Interestingly, when I use the Topaz model strictly for deinterlacing without any upscaling (keeping the original 720x576 resolution) and select Original Pixel Type, everything works perfectly. The output goes back into Resolve with the correct aspect ratio and no distortion. The issue only appears when the upscaling process is introduced.

My Questions:

Is this behavior typical or known when handling anamorphic/non-square PAL material in a Resolve-to-Topaz roundtrip?

Is there a clean workaround directly within Topaz (or specific export settings in Resolve) to avoid having to manually tweak the width in post?

What are your best practices or experiences when converting non-square SD assets to square-pixel UHD within this kind of pipeline?

Looking forward to your insights!

Best regards,

Robert

2

I believe I responded to your support ticket about this as well earlier this morning. As I mentioned in the reply there are some issues the team has been working on trying to address from videos with odd metadata tags, that include different values for PAR, DAR and SAR on older video content. As they can get things sorted out and add in further support, or new workflows we will make a note in the release changelogs.

I will leave this open, though, here to allow other users to share their workflows if they wish.

3

I use Hybrid to deinterlace and set correct aspect ratio of the output: 768x576.

This way Topaz will work with non-distorted progressive output.

Here’s the detailed explanation of my Hybrid workflow. Don’t be afraid of the settings, it’s one time setup.