Topaz Deinterlace

Old DVD rips, TV captures, miniDV transfers, any footage recorded on interlaced video where every other horizontal line belongs to a different moment in time. Topaz Deinterlace runs Topaz's Dione models on that material and unscrews the combing - and it can upscale in the same pass, since interlaced sources are usually SD and you want them bigger anyway. If you've got a box of archive footage, this node plus the rest of the restore chain is the closest thing to a one-stop deinterlace-and-enhance pipeline you'll find in ComfyUI.

What it actually does

The node is a thin, constrained wrapper over the same tvai_up filter the big Upscale node uses - but as a dedicated node the purpose is visible and the constraints are enforced rather than discovered through a failed render. Two things are enforced here:

• The model list is filtered to Dione only (ddv/dtv/dtd families), so you can't accidentally pick a generic upscaler and wonder why the combing survived. The tooltip on model tells you which Dione variants want what: Dione DV/TV/Dehalo expect genuinely interlaced input, while the Dione Robust variants take progressive input (their job is sharpening already-deinterlaced or progressive footage).
• scale_factor is 1, 2 or 4. 1 is deinterlacing alone; higher values deinterlace and upscale in one pass, which is both faster and usually the right call for SD sources. fps defaults to 25 because interlaced material is usually 25 or 29.97 - set it to match your source.

Output is a single images IMAGE batch, resolution scaled by your factor.

Why there's no field-order control

Look closely and you'll notice there's no top-field-first / bottom-field-first dropdown. That's deliberate, and the story is worth knowing because it's the difference between a pack that's honest and one that ships dead weight. It used to have that control, sending a model parameter called interlacing - which did nothing. Topaz's own filter documentation lists only three parameter groups (Hyperion, SAM2 and Grain), no interlacing, and Dione models take no parameters at all. Because parameters is a dictionary option, ffmpeg accepts any key you throw at it and silently drops the ones it doesn't recognise. On genuinely interlaced test footage both field-order settings produced identical output.

So the author measured it, found the control was theatre, and removed it rather than leave a knob that does nothing - on the reasoning that a control which silently has no effect is worse than none. The Dione models figure out field order themselves, and on material combed 210x above the progressive floor, ddv-3 removed 72% of the combing while motion stayed at the right speed.

Install

It ships with the whole pack - from ComfyUI's custom_nodes folder:

git clone https://github.com/donangel85/ComfyUI-TopazVideoLocal.git

Restart ComfyUI; nothing to pip install (numpy is already present). ComfyUI Manager finds it by searching "TopazVideoLocal". Underneath: Windows and a licensed, signed-in Topaz Video installation with the Dione weights on disk.

Common issues

Run the pack's Topaz Diagnostics node before anything else - it tells you whether the install, models or licence is the blocker. If your source is progressive and you run a Dione variant that expects interlaced input, you'll get a weird result; that's what the Dione Robust models are for. And the honest advice from the pack's own testing: deinterlacing progressive footage doesn't help it, it just costs render time - this is a repair node for footage that actually has combing.