EmptyLatentImage64

EmptyLatentImage64 is about as simple as a node gets: it makes an empty (all-zero) latent at the width, height and batch size you set, to start a text-to-image generation from. If you've used ComfyUI's built-in EmptyLatentImage, this is the same idea - a blank canvas for the sampler to denoise into a picture. RES4LYF ships its own so its example workflows are self-contained and don't lean on a node from another pack.

There's not much mechanism to explain. Every diffusion generation starts from a latent, and for txt2img that latent is just noise-to-be - an empty tensor of the right dimensions. This node produces one. The reason to use it over the stock version is purely convenience: it's already there when you install RES4LYF, and it drops straight into the pack's sampler nodes. If you need something fancier - a specific channel count, Stable Cascade geometry, or a custom precision - that's what the pack's EmptyLatentImageCustom is for. For a normal SD/SDXL/Flux txt2img start, this is all you need.

How it works

You set the output size and batch, it allocates a zero-filled latent of that shape and hands it out. The sampler adds noise and denoises from there.

The inputs and outputs that matter

Three widgets, one output - nothing to overthink:

• width / height (default 1024) - the image dimensions in pixels. The node converts to the model's latent resolution internally, same as the stock node.
• batch_size (default 1) - how many images to generate at once. Bump it to explore several seeds in one run.
• latent (LATENT) out - wire it into your sampler's latent input.

How to install it

ComfyUI Manager: search RES4LYF, install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/ClownsharkBatwing/RES4LYF
cd RES4LYF
pip install -r requirements.txt

then restart and hard-refresh (F5). Nothing to download.

Common issues & troubleshooting

Odd dimensions give odd results. Diffusion models want dimensions that divide cleanly into the latent grid - stick to multiples of 8 (the widget steps by 8 for this reason), and to resolutions your model was trained near (around 1024×1024 for SDXL/Flux, 512-ish for SD1.5). Wildly off-spec sizes produce distortion, and that's the model, not the node.

I need a non-standard latent shape. Then this isn't the node - use EmptyLatentImageCustom, which exposes channel count, compression and precision for Cascade and 16-channel models.

Which empty-latent node should I use? For everyday txt2img, either this or the stock EmptyLatentImage is fine; they do the same job. Use this one to keep a RES4LYF workflow tidy and self-contained.