Topaz Upscale Params

Every upscale model in the ComfyUI-TopazVideoLocal pack has a bag of tuning parameters - noise removal, detail recovery, halo reduction, compression cleanup - and this node is where you set them. It produces a params bundle you feed into the params socket of any upscale-capable node (video, image, even the deinterlace and motion-deblur wrappers), so you can build one tuning profile and reuse it across a graph instead of re-setting sliders per node.

The honest advice for a beginner: don't touch the sliders at all on your first runs. Topaz's models at 2x with default tuning already beat a plain Lanczos resize by a wide margin (the pack measured 28.56 dB / SSIM 0.928 against Lanczos's 27.75 / 0.919 on real footage). The node exists for when default isn't good enough - noisy source, compressed source, AI-generated video, archive film - and that's what the profile dropdown is for.

The two ways to use it

profile (default manual) is the top-level switch. Leave it on manual and the sliders below apply as usual. Pick anything else and that profile's values get used. Three kinds of entry appear and the prefix tells them apart:

• Topaz: … - read live from Topaz Video's own preset folder, authored by Topaz Labs (only the ones that actually set tuning values; the rest just pick a model and output settings).
• My: … - presets you've saved yourself.
• No prefix - starting points shipped with this pack, like Compressed / web video, Old film / archive, AI-generated video. Useful defaults, not official Topaz values.

Then profile_strength scales whatever profile you picked - 0.5 is half as aggressive, 0 disables its tuning entirely. Change it before picking the profile, or use the reload button after.

The sliders that matter

All run -1 to 1 and the tooltips are the author's own descriptions, so trust them: noise removes ISO noise (too high also removes fine detail), details recovers texture lost to in-camera noise suppression, halo reduces ringing around strong edges from oversharpening, blur adds sharpening for soft sources, compression reduces blockiness from codecs, preblur goes negative for aliasing/moire or positive for lens blur. Beyond those, grain (with grain_size and grain_type) adds film grain, blend mixes the original back in, and prenoise adds noise before processing - all optional, all zero by default, and all easy to overdo.

Two more worth naming: color_correction (on by default) does extra colour correction where the model needs it, and auto_estimate_frames analyses up to N frames to estimate the parameters automatically - 0 disables it. That's the same estimation family as the separate Topaz Parameter Estimate node, inline.

The presets-in-the-browser bit

The node ships a small frontend extension. In the browser, picking a profile copies its values into the sliders so you can see what it actually does and adjust - and edit_preset_values (the last widget on the node) decides who wins: off means the preset runs exactly as authored and the sliders are ignored; on means your edited sliders count. It defaults to off, but picking a profile in the browser turns it on for you, because the sliders now hold that preset's values and adjusting them is the point. Two buttons round it out: Reload preset into sliders (discard an experiment and start over) and Save sliders as preset (stores the current values as My: <name>, kept in user_presets.json beside the package). Through the plain API, without the browser extension, the dropdown behaves like a normal server-side control and edit_preset_values stays off.

Install

It ships in the pack:

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

from ComfyUI's custom_nodes folder, restart. No pip step beyond numpy; ComfyUI Manager finds it under "TopazVideoLocal". Real requirements underneath: Windows, a licensed signed-in Topaz Video installation.

Common issues

The classic confusion is adjusting sliders and seeing no change - that's edit_preset_values sitting in the "preset as-is" position, so the profile wins and your edits are ignored. The second is mistaking a shipped starting-point profile for an official Topaz value; the log line after each run shows exactly what got resolved, so check it if a result looks more aggressive than you intended.