ZIT Empty Latent Image MXD

Z-Image Turbo is the Apache 2.0 image model that quietly became infrastructure - the one people reference in passing as the base for a workflow rather than post about as a topic. It's also got opinions about resolution: every dimension has to be a multiple of 64, and it behaves best under a ~6.5MP ceiling. Get the size wrong and you're in silent-quality-loss territory. ZIT Empty Latent Image MXD is the preset latent generator that keeps you on the right side of those rules without thinking about them.

What it is

The Z-Image member of the MaxedOut pack's resolution-preset family - same idea as the Flux, Flux 2, and SDXL versions: pick a resolution, flip vertical if you want portrait, set a batch_size, get a LATENT. But the presets are tuned specifically to Z-Image's constraints, which is where the value lives.

The dropdown is organized in three tiers:

• High: Square 1536², Photo 1792x1344, Landscape 1920x1280, Widescreen 2048x1152, Ultrawide 2304x1024
• Standard: Square 1024², Photo 1152x896, Landscape 1280x832, Widescreen 1344x768, Ultrawide 1536x640
• Low: Square 512², Photo 576x448, Landscape 640x448, Widescreen 704x384, Ultrawide 768x320

Every entry is a multiple of 64, and the High tier stays under the 6.5MP ceiling.

How it works

Two hard guards baked into the generator:

1. The rule of 64. If a resolution isn't a multiple of 64 in either dimension, it raises - the source comments call this out as a Z-Image training constraint, and the node enforces it rather than letting you slide.
2. The pixel ceiling. Presets must stay at or below 6.5MP total, matching where the model behaves.

On valid input it allocates the empty latent at [batch, 16, h//8, w//8] - Z-Image uses the 16-channel / 8x-compressed latent it shares with the SD3 family - and returns it as a standard {"samples": ...} latent.

Inputs and outputs

• resolution - the dropdown, default Square (1:1) 1024x1024 (the 1MP baseline).
• vertical - swap to portrait.
• batch_size - 1 unless you're batching.
• Output: LATENT → sampler's latent input.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Maxed-Out-99/ComfyUI-MaxedOut.git

Restart, or ComfyUI Manager → search "Maxed Out" → install. No model files, no extra deps.

Troubleshooting

• "Invalid preset … Z-Image Turbo requires multiples of 64" - the guard fired. Every preset in the dropdown is valid, so this only happens if you're feeding values through some automation. Stay on the presets.
• "presets must stay at or below 6,500,000 pixels" - same story; the dropdown is safe by construction.
• Wrong latent shape for your model - this makes 16-channel latents for Z-Image. If you're actually sampling a different model, use its matching latent node. The dropdown says ZIT, and it means it.

For "I want Z-Image Turbo output that isn't quietly degraded by a bad size", this is the no-thought answer - a two-input node that removes an entire class of failure.