❌ (OLD) Latent Settings (Flux/SDXL)

The display name says it all: ❌ (OLD) Latent Settings (Flux/SDXL). This is the earlier version of Light-x02's latent-size node, and the author has marked it deprecated in favor of the 🧱 Universal Latent Settings that ships alongside it. If you're starting fresh, skip this one and use Universal Latent Settings. If you inherited a workflow that still uses it, here's what it does and why you might not want to stay on it.

What it is

It's a "make me an empty latent at the right size" node - the same job as ComfyUI's core Empty Latent Image, but with curated resolution lists for Flux and SDXL plus a toggle to switch between them. You pick a resolution from the Flux list or the SDXL list, flip a mode_resolution boolean (on = Flux, off = SDXL), and it produces a blank LATENT tensor plus the resolved width and height as integers.

The outputs matter more than they look: width and height let you wire the chosen resolution into other nodes (like a resolution-aware conditioning or metadata text) without parsing strings yourself.

The mechanism

An empty latent is just a tensor of zeros shaped [batch, 4, height/8, width/8] - 4 latent channels, 8× downsampled per side, the SDXL-style layout. The node picks the base size from whichever resolution list is active, applies flip_orientation if set, and lets width_override/height_override (0 = off) override the preset entirely. batch_size controls how many blank latents you get.

The one caveat worth internalizing: the latent it builds is always 4-channel. Flux and Z-Image run on a 16-channel VAE, so a 4-channel latent won't directly feed their samplers - for those models, treat this node as a size calculator and use the width/height outputs with ComfyUI's EmptySD3LatentImage instead. For SDXL and its finetunes, the 4-channel latent is exactly right.

Why it got retired

Universal Latent Settings fixes the two annoyances of this one: it adds the Z-Image resolution family, and - more importantly - it replaces the two separate resolution dropdowns with a single dropdown that updates dynamically based on the model family you select (the JS-driven UI). Managing two parallel lists and a mode toggle was error-prone; one list filtered by model is the better design. It also got the bug fixes, like the "generates at 1536×1536 even when you selected something else after switching workflows" fix noted in the changelog.

Inputs / outputs at a glance

• mode_resolution - Flux (on) vs SDXL (off).
• resolution_flux / resolution_sdxl - the two curated lists, 26 entries each, labeled with aspect ratio.
• flip_orientation, batch_size, width_override, height_override - the knobs.
• Outputs: LATENT, width, height.

Install

Ships in Light-x02 Nodes - ComfyUI Manager → search "ComfyUI-Lightx02-Nodes", or:

cd ComfyUI/custom_nodes
git clone https://github.com/Light-x02/ComfyUI-Lightx02-Nodes

Restart ComfyUI. No extra dependencies, no models to download. If an old workflow breaks on a missing class, check whether you've been running the standalone Latent Settings fork from another repo - the pack README warns that leftover standalones cause duplicate-registration conflicts.