💪TB | Empty Latent Image

Every diffusion workflow starts with an empty latent - a tensor of noise waiting to be denoised into an image. ComfyUI's built-in EmptyLatentImage makes you type width and height. 💪TB | Empty Latent Image is the TinyBreaker version, and its party trick is that it doesn't ask you for dimensions at all: it reads the image size, aspect ratio, and batch count straight out of the GENPARAMS blob flowing through the pack. You set "16:9, large, batch of 2" once in your prompt or params chain, and this node just obeys.

How it works

Feed it a genparams and it pulls the image geometry from the chain's settings - width and height derived from the aspect/size you chose upstream, plus the batch size. It then divides those pixel dimensions by the VAE's block size (8×) to compute the latent resolution and returns a zeroed latent batch at that size. No widgets, no typing, no mismatch between "what I asked for in the prompt" and "what the latent actually is." The tooltip on the input is even a little funny - it literally says "The generation parameters ???" - which is a good hint that this node lives or dies by what the rest of your chain decided.

The one input, the one output

• genparams - the generation parameters. If this is the first genparams node in your chain, plug it after whatever seeds the blob (the Unified Prompt Input, the checkpoint loader's METADATA output, or a style node).
• Output - LATENT, the empty latent batch, ready for the sampler's latent input.

That's the entire surface. If you're used to vanilla ComfyUI, it feels almost too empty - but that's the point. The pack has consolidated image-geometry decisions (aspect, orientation, size, batch) into the genparams layer so you never have to keep a latent size in sync with your prompt settings by hand.

Install

Nothing special: ComfyUI Manager → search "tinybreaker", or

cd ComfyUI/custom_nodes
git clone https://github.com/martin-rizzo/ComfyUI-TinyBreaker

then restart. No model files needed for this node specifically.

Gotchas

The obvious failure mode: wire it up without a proper genparams upstream and the size defaults aren't what you expect - remember it's reading whatever's in the blob, so if your prompt's --aspect 16:9 isn't reflected in the output, the aspect isn't reaching this node's genparams (check your chain, not the node). It also only creates empty latents; it won't preserve or reshape anything you feed it, so it's strictly a starting point, not a size converter. And it replaced the deprecated Set Image node's job - old workflows that set aspect/size through that node expect the same values, but now they come from the genparams pipeline. If you're porting an old workflow, keep the same genparams order: seed and geometry setters should run before this node, sampler/refiner tweaks after.