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.