AC_Super_EmptLatent

AC_Super_EmptLatent is a drop-in replacement for ComfyUI's built-in Empty Latent Image, with two upgrades: a big dropdown of common resolutions and a couple of handy INT outputs. If you've ever typed 768, 1024, 1216 into that node by hand, you'll appreciate having the presets.

How it works

The mechanism is boring in the best way - torch.zeros([batch_size, 4, height // 8, width // 8]). That's the same all-zero latent tensor the core node makes: four channels because that's the VAE's latent shape, and both dimensions divided by 8 because that's how much the VAE downsamples. Zero-initialized latents are what "generate from scratch" means, so this node just makes the canvas you sample onto.

The inputs

• Boolean (Custom / All_ready) picks where the size comes from. Custom uses your width and height widgets; All_ready ignores them and parses the Resolution preset string instead.
• Resolution - 37 presets, from 512*512 up to cinematic 6400*4800. The useful mid-list ones like 1024*1024, 768*1280, and 1024*1536 are the SDXL-era favorites.
• width / height - used in Custom mode, 64 to 5277, step 1.
• batch_size - 1 to 64. Batch here means n independent latents stacked, which samples n images in one run.

The outputs

LATENT, plus WIDTH and HEIGHT as INTs. The last two are the sleeper feature: you can wire them into anything that needs to know the canvas size - a resize node, a save path, a comment - without typing the number twice. It's the "one source of truth" idea from the plumbing layer, applied to resolution.

Installing it

Part of ComfyUI_AC_FUNV8Beta1 by Cc啊程 (Bilibili ComfyUI teacher; the README is Chinese-only). ComfyUI Manager (search "AC_FUN") or:

cd ComfyUI/custom_nodes
git clone https://github.com/A719689614/ComfyUI_AC_FUNV8Beta1

Restart ComfyUI. No downloads.

Where people get burned

Not much to trip on here, but one thing: in All_ready mode the width/height widgets are dead - they look editable but the preset wins. Change the preset, not the numbers, or you'll think you resized and nothing happened. And a tip from the wider workflow: the WIDTH/HEIGHT outputs pair nicely with this pack's AC_Super_KSampler if you want to guarantee the sampler uses the exact same canvas.