LatentUpscaleMultiply

LatentUpscaleMultiply is a ghost from the pack's past. Before the Quality of Life Suit consolidated everything into one extension, latent scaling lived in its own standalone file - and this node is that old implementation, kept around so workflows built on the original survive. You can still search for it under O/deprecated/latent, and it still works, but it's the one you should actively replace.

The story, quickly: the pack's installer even hunts for and archives the legacy standalone files (LatentUpscaleMultiply_Omar92.py and friends) on startup, zipping them into a backup so they can't clash with the merged pack. This deprecated node is the same code living on inside the merged extension. The modern replacement is LatentUpscaleFactor _O in O/latent - same idea, slightly better behaviour.

How it works

Mechanically it's almost identical to the modern node: read the latent's width and height, multiply by WidthMul and HeightMul, hand the result to common_upscale with the chosen method and crop. The input names are the giveaway of its age: WidthMul / HeightMul instead of WidthFactor / HeightFactor.

Inputs:

• samples (LATENT)
• upscale_method - nearest-exact, bilinear, or area
• WidthMul (FLOAT) - width multiplier, default 1.25, range 0–10
• HeightMul (FLOAT) - height multiplier, default 1.25
• crop - disabled or center

One output: LATENT.

Where it differs from the modern node

Two small things, both in your favour if you switch:

1. The multiple-of-8 enforcement is slightly sloppier here. The old code rounds up to the next multiple of 8 unconditionally (d += 8 - leftover); the modern version rounds to the nearest multiple. The old way can push a dimension up a full 7 pixels when rounding down would have been closer. Minor, but it's the kind of difference that shows up in odd final sizes.
2. No 8192 cap. The deprecated node lacks the MAX_RESOLUTION clamp the modern one has, so a big factor on a big latent can produce a surprisingly large tensor.

Neither difference will wreck your workflow - that's why it still runs - but there's no upside to it either.

The honest advice

If you load an old workflow and this node appears, leave it be and let the graph run; it'll produce a valid latent. But if you're building anything new, don't search for it - grab LatentUpscaleFactor _O (or LatentUpscaleFactorSimple _O) from O/latent. Same one-line purpose, better rounding, a safety cap, and no "deprecated" tag hanging over it. And remember the general rule with latent upscaling in this pack: it's the cheap intermediate step of a two-pass workflow, not a true upscaler - keep factors modest and follow with a low-denoise second pass for detail.

Installing it

It comes with the Quality of Life Suit pack (omar92/ComfyUI-QualityOfLifeSuit_Omar92). Install via ComfyUI Manager (search "Quality of Life Suit") or clone:

cd <ComfyUI>/custom_nodes
git clone https://github.com/omar92/ComfyUI-QualityOfLifeSuit_Omar92.git

Restart ComfyUI properly. No models, no dependencies. On startup the pack may also archive any legacy standalone node files it finds in custom_nodes - if you had the old separate files installed, they'll be zipped up and removed, which is the pack cleaning house.

Common issues

• "I searched for this and found the 'Factor' node instead." Correct - that's the one you want.
• "My output size is a bit larger than expected." The old unconditional round-up to a multiple of 8. Switch to the modern node if it bothers you.
• "Is it broken?" No, it works. It's just surpassed.

It's a working relic, and the honest summary is one line: if you see LatentUpscaleMultiply, use LatentUpscaleFactor _O instead.