CustomLatentImageSimple(NYJY)

If CustomLatentImage-NYJY is the full-featured blank-canvas node, this is the stripped-down twin: same 29 SDXL ratio presets, but no overrides, no upscale factor, no extra size outputs. You pick a ratio, set a batch size, and get one LATENT out. That's the whole node. It exists for the workflows where the ratio dropdown is the only knob you ever touch, and where the extra outputs would just be clutter on the canvas.

How it works

Identical mechanism under the hood: the ratio string (13:19 - 832x1216, default) maps to pixel dimensions, and the node creates a zero-filled latent tensor at [batch, 4, height/8, width/8] - the standard 8× VAE downsampling that every SDXL-family model expects. The sampler denoises this blank slate into your image. Nothing is downloaded, nothing is configured; it's pure tensor creation.

Inputs and outputs

• ratio - the 29-preset dropdown, same list as the full node. Default 13:19 - 832x1216.
• switch_width_height - flips the ratio's orientation (portrait ↔ landscape) for when you want the same ratio reversed.
• batch_size - number of blank latents, default 1.

One output: LATENT. That's all.

Installing

Standard pack install: ComfyUI Manager → Install via Git URL → https://github.com/aidenli/ComfyUI_NYJY, restart.

Which one should you use?

Honest answer: if you only ever generate at a preset ratio, this one keeps your graph clean - one less output wire, no temptation to start fiddling with overrides. If you ever scale up for a high-res pass, switch to the full CustomLatentImage-NYJY node, because its upscale_factor and the resolved width/height outputs save you from building that math yourself with separate nodes. Same ratios, same SDXL-family caveat (Flux and other non-SDXL architectures have their own preferred resolutions), just one is the economy version and the other the deluxe. For the simple case, simple wins.