2026-06-20 23:18:27 -05:00
2026-06-20 23:18:27 -05:00
2026-06-20 23:18:27 -05:00
2026-05-10 20:36:50 -05:00
2026-05-10 20:36:50 -05:00
2026-05-10 20:36:50 -05:00
2026-06-20 23:18:27 -05:00
2026-06-20 23:18:27 -05:00

ComfyUI Latent Presets

Model-scope latent presets for ComfyUI

Important

This extension uses the comfy_api.latest v3 node API. It requires a recent ComfyUI build (0.3.x or later). Older installations will not load these nodes.


Installation

Via ComfyUI-Manager — use the GitHub address to directly add the extension since it is not currently registered.

Manual

cd ComfyUI/custom_nodes
git clone https://github.com/bnwa/ComfyUI-LatentPresets

No additional Python dependencies are required beyond ComfyUI itself.


Nodes

All nodes appear under the models/latent category.

Select SDXL Dimensions

Picks from nine community-standard SDXL dimension presets and emits a pre-allocated empty latent alongside scaled output dimensions.

Name Type Default Range
Input dimensions Combo — 9 presets (see below)
Input upscale_factor Float 2.0 1.0 – 8.0, step 0.25
Input batch Int 1 ≥ 1
Name Type
Output width Int
Output height Int
Output latent Latent (4-channel)
Output upscale factor Float
Output upscaled width Int
Output upscaled height Int

Presets

Label Width Height Ratio
1024 × 1024 1024 1024 1:1
1152 × 896 1152 896 9:7
896 × 1152 896 1152 7:9
1216 × 832 1216 832 19:13
832 × 1216 832 1216 13:19
1344 × 768 1344 768 7:4
768 × 1344 768 1344 4:7
1536 × 640 1536 640 12:5
640 × 1536 640 1536 5:12

Select Flux.1 Dimensions

Picks from 18 Flux.1 Dev dimension presets across two megapixel budgets and emits a pre-allocated empty latent using the 16-channel Flux VAE latent space.

Name Type Default Range
Input dimensions Combo — 18 presets (see below)
Input upscale_factor Float 2.0 1.0 – 8.0, step 0.25
Input batch Int 1 ≥ 1
Name Type
Output width Int
Output height Int
Output latent Latent (16-channel)
Output upscale factor Float
Output upscaled width Int
Output upscaled height Int

Presets — ~1 MP (community-standard)

Label Width Height
Square 1024 1024
Landscape 1152 896
Portrait 896 1152
Classic Landscape 1152 864
Classic Portrait 864 1152
Photo Landscape 1216 832
Photo Portrait 832 1216
Wide Landscape 1344 768
Tall Portrait 768 1344
Widescreen 1280 720
Vertical 720 1280
Ultrawide 1536 640
Ultra-Tall 640 1536

Presets — ~2 MP (community sweet spot)

Label Width Height
HD Square 1408 1408
HD Landscape 1728 1152
HD Portrait 1152 1728
Full HD Widescreen 1920 1088
Full HD Vertical 1088 1920

Select Z-Image Dimensions

Picks from 33 official Z-Image dimension presets across three resolution tiers and emits a pre-allocated empty latent using the 16-channel VAE latent space (8× spatial compression).

Name Type Default Range
Input dimensions Combo — 33 presets across three tiers
Input upscale_factor Float 2.0 1.0 – 8.0, step 0.25
Input batch Int 1 ≥ 1
Name Type
Output width Int
Output height Int
Output latent Latent (16-channel)
Output upscale factor Float
Output upscaled width Int
Output upscaled height Int

Tiers: 1024 (~1 MP, 11 presets) · 1280 (~1.6 MP, 11 presets) · 1536 (~2.4 MP, 11 presets). Each tier covers Square, Landscape/Portrait, Classic 4:3/3:4, Photo 3:2/2:3, 16:9/9:16 Widescreen/Vertical, and 21:9/9:21 Ultrawide/Ultra-Tall.

S
Description
No description provided
Readme GPL-3.0
81 KiB
Languages
Python 100%