ResolutionMultiplySlider

Flux Continuum's upscaling module is controlled by a simple idea: generate at native resolution, then multiply. That's exactly what this slider does. It's a float slider from 1 to 10 in 0.1 steps, default 1.0, and its description is refreshingly literal: "controlling resolution multiplication for upscaling." Move it to 2 and the image resolution doubles. That's the whole mechanism - one value out, and the upscaling section of the workflow multiplies the processed image's resolution by it.

Why a multiplier and not a resolution picker

The Continuum workflow already has a Resolution Picker for choosing the generation size, with pre-calculated Flux-friendly options around 1024 base. This slider is a separate lever for what happens after - the README calls it "Resolution Multiply: multiplies image resolution after any preprocessing." So the two controls don't fight: you pick your native generation size up front, then decide how hard to push the upscale at the end.

It's worth being clear about what this does and doesn't buy you. Multiplying resolution adds pixels. Whether those pixels carry real detail is up to the upscale model you've loaded (the README recommends 4xNomos8kDAT), and whether they carry new detail is up to the generative side of your upscaler. Pure multiplication is cheap and can't hallucinate, which is the "more pixels" job in the upscaling toolbox. If you want invented detail, that's a different rung - a detailer pass or a generative upscaler - and no amount of multiplier is a substitute.

The numbers you should actually care about

Keep an eye on the square in the name: it's a linear multiplier. At 1.5 your linear dimensions go up 50%, which is 2.25x the pixels. At 2.0 you're quadrupling the pixel count. Starting from a 1024 base, 2x is already a 2048-wide image - well past Flux's native comfort zone, and the VAE decode plus upscaler will feel it in VRAM. Don't drag it to 10 on a 1024 source and expect anything but an out-of-memory error or a very long wait. Most real work lives between 1.5 and 2.5.

Also worth knowing: Flux does not like being generated directly at huge resolutions - tiling and repeated anatomy kick in past its native range (the troubleshooting lore is loud about this). The sane path is exactly what this workflow does: generate at ~1024, then multiply and upscale after.

Install and wiring

Ships with the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/robertvoy/ComfyUI-Flux-Continuum

Restart, or install via ComfyUI Manager → search "Flux Continuum". The node itself is dependency-free; the upscaling module it feeds will want an upscale model file in your upscale_models folder.

One FLOAT out, wired into the upscaling section. That's it. If the slider is greyed or the upscale does nothing, the model isn't loaded or the wire is detached - check the Upscale Model selector in the workflow's config panel before you suspect the slider.

Troubleshooting

• OOM on upscale - your multiplier is too high for your VRAM. Drop it, or tile the upscale.
• Soft/blurry upscale - that's the upscale model, not the slider. Switch the model in the config panel; 4xNomos8kDAT is the pack's recommended default for a reason.

It's one knob, it does one job, and getting it wrong costs you VRAM rather than quality. Treat it like a speedometer, not a tuning knob.