DARASK Empty Latent (Preset)

Every model generation has the same dull step: build an empty latent at a resolution that actually matches your subject. The stock EmptyLatentImage makes you type width and height as raw numbers, and nobody remembers whether 832×1216 is the portrait or the landscape one. DARASK Empty Latent (Preset) replaces those two numeric widgets with a single dropdown of labeled presets - each one showing its pixel dimensions and its aspect ratio, because "832 × 1216 (13:19 ≈ 2:3 portrait)" is a lot easier to keep straight than the bare digits. It's a tiny quality-of-life swap, and once you use it you'll resent typing resolutions again.

How it works

There's no magic here - it's an EmptyLatentImage with nicer controls. The eight presets cover the standard 1:1, portrait and landscape targets, including the 2x square and the wide 4:7 / 7:4 slots:

• 1024 × 1024 (1:1) and 2048 × 2048 (1:1, 2x)
• 832 × 1216 (≈2:3 portrait) and 1216 × 832 (≈3:2 landscape)
• 896 × 1152 (≈3:4 portrait) and 1152 × 896 (≈4:3 landscape)
• 768 × 1344 (4:7 portrait, wide) and 1344 × 768 (7:4 landscape, wide)

swap_orientation flips width/height in one click - the fastest way to get from the portrait to the landscape version of the same ratio.

The input that does real work

batch_size looks boring, but it's the node's hidden superpower. Right-click the widget → Convert Widget to Input, then wire it to the total_count output of the DARASK Prompt Cell Output node. Now your latent batch size automatically tracks how many prompt combinations your chain produced - add a variant line upstream and the batch grows to match, no manual sync ever again. That single connection is the whole "generate every prompt combination in one queue" workflow the pack is built around. Outputs are latent (into your sampler) plus width and height as integers you can log or feed elsewhere.

Install

Standard DARASK pack: ComfyUI Manager → search DARASK, install, restart. No dependencies beyond the pack itself, no models. This is one of the pure-utility nodes in the pack - it depends on nothing heavy, so it's a safe early install.

When it's not enough

If you need a resolution that isn't in the eight presets, this node has no free-form escape hatch - you'll want a plain EmptyLatentImage for that. And 2048×2048 at any meaningful batch size is a VRAM wall on mid-range cards; batch swells combine badly with the 2x preset, so be deliberate with total_count when it's wired to batch size.