Ethanfel
291a0a1f1c
Splits a [B,C,H,W] latent into left/right or top/bottom halves, or four quadrants. Model-agnostic (SD, SDXL, Flux / FLUX.1 Krea); preserves batch dimension and noise_mask. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
1.6 KiB
ComfyUI-Latent-splitter
A ComfyUI custom node that splits a latent into 2 (left/right or top/bottom) or 4 tiles along its spatial axes.
It is model-agnostic — it only slices the height/width of the [B, C, H, W]
latent tensor, so it works with SD/SDXL (4-channel) and Flux / FLUX.1 Krea
(16-channel) latents alike. A 1024×1024 image (a 128×128 latent) splits into
two 512×1024 halves or four 512×512 quadrants.
Node: Latent Split (2 / 4)
Category: latent
Inputs
| Input | Type | Notes |
|---|---|---|
samples |
LATENT | The latent to split. |
mode |
choice | left / right, top / bottom, or quad (4). |
Outputs — filled in reading (row-major) order:
| Mode | latent_1 | latent_2 | latent_3 | latent_4 |
|---|---|---|---|---|
left / right |
left | right | — | — |
top / bottom |
top | bottom | — | — |
quad (4) |
top-left | top-right | bottom-left | bottom-right |
Unused outputs (3 and 4 in the two-way modes) return None — leave them
unconnected. Odd latent dimensions split with the extra row/column going to the
second tile. The batch dimension and any noise_mask are preserved (the mask is
sliced to match each tile).
Install
Clone into your ComfyUI custom_nodes directory:
cd ComfyUI/custom_nodes
git clone https://github.com/ethanfel/ComfyUI-Split-Latent.git
No extra dependencies — it uses the PyTorch that ships with ComfyUI. Restart ComfyUI and find the node under Add Node → latent → Latent Split (2 / 4).