Topaz Motion Deblur

Fast action shot at the wrong shutter speed. A pan that smeared the whole frame. Handheld footage where the subject moved and the camera moved and both of them smeared. Topaz Motion Deblur runs Topaz's Themis model over your IMAGE batch to pull motion blur back out of footage - on your local Topaz Video installation, same raw-frame pipe as the rest of this pack.

But here's the part to internalize before you wire it in everywhere: this is a repair node, not a sharpener. The pack's own measurements on real footage are unusually blunt about it. Themis gives back about a fifth (18%) of the detail a genuine motion blur destroys - real recovery, not a sharpening illusion. But run it on footage that was already sharp and it costs roughly 8% of the picture's gradient energy. It's not free, even when it has nothing to fix. Put it in the graph where there is motion blur to remove, and bypass it otherwise. That's the entire skill with this node.

What it does (and doesn't)

Themis only supports scale 1. Resolution never changes - there's no scale factor widget because there's nothing to offer; a control that can only hold one value isn't worth a dropdown. If you want both deblurring and more resolution, you chain a Topaz Video Upscale node after it. That's the intended shape of the restore workflow: repair passes first, resolution last.

Inputs are just images, model (filtered to Themis variants, thm-2 as the sensible default), and fps. Optional params accepts tuning from a Topaz Upscale Params node if you want to nudge it, and engine if you're overriding engine settings. Output is a single images batch at the input resolution.

Where it sits in a restore chain

The pack ships a restore example that runs stabilize, deinterlace and motion deblur side by side on the same source, plus a chained restore workflow, and the ordering is the thing that makes sense: stabilise first (deblur can't fix a moving frame it can't register), deinterlace if the source is interlaced, then deblur, then upscale last. If you only have one blurry clip, you don't need all of that - you need this node between your source and your upscale.

Install

Same pack as the rest:

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

from ComfyUI's custom_nodes folder, then restart ComfyUI. No pip step beyond the numpy every ComfyUI has; ComfyUI Manager finds it by searching "TopazVideoLocal". Real requirements: Windows, a licensed signed-in Topaz Video installation, and the Themis weights on disk - [download required] on a model means letting Topaz fetch it via the Engine Settings allow_model_download, or running it once in the Topaz app.

Common issues

Start from the pack's Topaz Diagnostics node if a run fails - install, models and licence status in one read. If you can't tell whether it did anything, that's usually because the source was sharp to begin with and you paid the 8% for nothing. Compare a bypassed frame against the output on footage you know had blur, and leave the node out of graphs where the clip is already clean.