🔹 CYH Latent | Flux Aspect Ratio

Flux doesn't take kindly to arbitrary resolutions, and the community has quietly converged on a handful of sizes that actually work - 1344×768 for widescreen, 1024×1024 for square, 1152×768 for 3:2. This node is basically a lookup table for those. You pick a ratio from a dropdown and it hands you an empty latent at the "right" Flux resolution, so you never have to remember whether it was 1344×768 or 768×1344 again.

It's one of six aspect-ratio generators in the Chye ComfyUI Toolset pack, all built the same way. Where the built-in Empty Latent Image node makes you type numbers, these give you presets tuned to a specific model family - here, Flux (v1-dev and the .1/schnell era, where the 1024 base and its multiples were the sweet spot).

How it works

Under the hood it's embarrassingly simple, and that's a good thing. The dropdown label already bakes the resolution into the text - 16:9 (Widescreen) - 1344×768 - so the node parses the ratio, looks it up in a table, applies your orientation and multiplier, rounds both dimensions to a multiple of 32, then creates a zeroed tensor:

latent = torch.zeros([batch_size, 4, final_height // 8, final_width // 8])

The // 8 is the VAE downsampling factor: you want pixel dimensions divisible by 32 so the latent ends up clean at /8. All you get out is a LATENT, exactly like ComfyUI's own Empty Latent Image - wire it into a KSampler's latent input and you're off.

The inputs that matter

Four inputs, and you'll mostly touch two of them:

• aspect_ratio - five presets: 1:1 (1024×1024), 4:3 (1280×960), 3:2 (1152×768), 16:9 (1344×768), 21:9 (1792×768). The resolution is right there in the label, which is the nicest thing about this node.
• multiplier - scales the preset by 0.1–10.0 before rounding. This is the knob to watch: 16:9 at multiplier 2.0 is a 2688×1536 latent, which is past where most Flux builds stay coherent in one pass. Keep it near 1.0 and upscale later if you want more.
• orientation - Portrait/Landscape. Defaults to Portrait, which is a bit of a trap: the preset already gives you a landscape resolution, and this setting just swaps width/height. Set it to Landscape for a widescreen image and it's a no-op.
• batch_size - 1–64, generates that many empty latents at once. Useful for grid runs, same as batch size on Empty Latent Image.

Where it sits in a workflow

CYH Latent | Flux Aspect Ratio → KSampler → VAE Decode is the whole trick. It replaces the classic dance of Empty Latent Image + math nodes when you want Flux's canonical ratios. Since the latents are all zeros, you can also feed one straight into an img2img or inpainting setup as a blank canvas.

Install

ComfyUI Manager (search "Chye ComfyUI Toolset"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/chyer/Chye-ComfyUI-Toolset
cd Chye-ComfyUI-Toolset
pip install -r requirements.txt

Restart ComfyUI after. The dependency list is scipy and opencv-python (the chunky ones) plus requests and coloredlogs; you already have torch and numpy. One quirk: the README insists the pack needs a .git/.cnr-id file containing Chye-ComfyUI-Toolset or workflow loads can error - the git clone handles it, but if you installed from a zip, create it.

Honest verdict

If you're already happy typing numbers into Empty Latent Image, this changes nothing. If you generate a lot of Flux at widescreen and are tired of the arithmetic, it's a small quality-of-life win. Just remember the presets are Flux.1-era - Flux 2 and Klein live in a 1MP–2MP native band, and these sizes (0.6–1.4MP) still land inside it fine.