🐟Preset Size Latent

Everyone has the same little ritual: you want a 16:9 wallpaper, so you type 1920 and 1080 into an EmptyLatentImage, or you want an Instagram Story so you go look up what size that actually is. PresetSizeLatent replaces the lookup with a dropdown - pick "Instagram Story," get 1080×1920, done - while still working like a normal empty-latent node for the sampler.

How it works

It's a drop-in replacement for EmptyLatentImage with a curated preset list bolted on. Flip use_preset on and pick from over 30 named presets covering portrait/landscape ratios, phone screens, video standards (4K, 1080p, cinema 2.35:1), print sizes, and specific platforms (TikTok, Instagram Story, YouTube cover, Xiaohongshu, Bilibili, and a few others). Flip it off and the width/height fields take over as a manual entry, same as the stock node. Either way it also exposes swap_dimensions to flip portrait/landscape without re-picking a preset, scale_factor to multiply the chosen size up or down, and batch_size for generating more than one latent at once.

If none of the presets fit, the README points out the list itself is just JSON - nodes/preset_sizes.json and nodes/camera_sizes.json inside the pack - so you can add your own by editing those files directly.

The inputs and outputs that matter

• use_preset (BOOLEAN, default on) - toggles between the preset dropdown and manual width/height.
• preset - the named size list; over 30 choices spanning social platforms, video, print, and general aspect ratios.
• width / height (INT, 64–8192, step 8, defaults 1080×1920) - used directly when use_preset is off, otherwise these get overridden by the preset choice.
• swap_dimensions (BOOLEAN) - flips whichever width/height you land on.
• scale_factor (FLOAT, 0.1–10, default 1) - multiplies the final dimensions.
• batch_size (INT, 1–64, default 1) - how many latents to generate at once.

Outputs: width and height (INT, handy for wiring into any downstream node that wants the raw numbers) and latent (LATENT, wire straight into your KSampler's latent_image input).

How to install it

Search ComfyUI-QHNodes in ComfyUI Manager, or:

cd ComfyUI/custom_nodes
git clone https://github.com/liuqianhonga/ComfyUI-QHNodes.git

Restart ComfyUI. This is one of the main-repo nodes, no submodules or extra downloads needed.

Common issues & troubleshooting

Non-multiple-of-8 dimensions. The width/height fields step by 8 for a reason - most VAEs (Flux and SD-family included) downsample the image by a factor of 8 into latent space, so dimensions that aren't multiples of 8 either get silently rounded or produce artifacts at the edges. The node enforces the step already, so this mostly bites if you're wiring in a custom width/height from somewhere else that bypasses the widget's constraint.

Preset dimensions don't match what you expected. Some presets are similarly named but different ratios (a handful of "portrait" and "mobile" variants sit close together in the list) - worth double-checking the actual pixel numbers shown in the dropdown label rather than trusting the name alone.

Editing the preset JSON and nothing changes. Custom presets require a ComfyUI restart to be picked up - the list is read once when the node loads, not live from disk on every run.