SharkOptions UltraCascade Latent

This one's for a specific model: UltraCascade, ClownsharkBatwing's Stable Cascade project. SharkOptions UltraCascade Latent sets the latent dimensions for a Cascade-style generation and packs them into the OPTIONS bundle a RES4LYF sampler consumes. If you're not running UltraCascade (or Stable Cascade / UltraPixel), you can skip this node entirely.

The reason the defaults look so small - width 60, height 36 - is that Stable Cascade works in an aggressively compressed latent space, far tighter than SD or Flux. Those little numbers correspond to a large final image once decoded. The README singles Cascade out as having "an excellent understanding of style," which is why it stays in the pack's orbit despite being an older architecture, and this options node is how you tell the sampler what canvas to work at in that compressed space.

How it works

You give it a width and height in Cascade's latent units, and it emits (or extends) an OPTIONS bundle carrying those dimensions. Chain it with the pack's other options nodes and route the result into a Shark/Clown sampler set up for UltraCascade.

The inputs and outputs that matter

• width (default 60) / height (default 36) - the latent dimensions in Cascade's compressed space, not pixels. Keep the aspect ratio you want in the final image; scale both together for a bigger output.
• options (optional OPTIONS) - chain another options node here to combine settings rather than start fresh.

Output is options (OPTIONS) - wire it into the sampler's options input or the next link in the chain.

How to install it

Part of RES4LYF. 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

Portable builds use the embedded pip. Restart, hard-refresh (F5). You'll also need the UltraCascade / Stable Cascade model files themselves - this node configures sampling, it doesn't download a model.

Common issues

The dimensions confuse everyone at first, because 60×36 isn't a resolution you recognize. Remember it's the compressed latent, so treat those numbers as a ratio-and-scale control and judge the result by the decoded image, not the input values. Second, this only makes sense inside a Cascade pipeline - dropping it into a Flux or SD3.5 graph does nothing useful, because the sampler isn't expecting Cascade-space dimensions. Third, the usual OPTIONS-chain rule: if the size seems ignored, confirm the output is actually wired into the sampler and not overwritten by a later options node. And as a "Beta" node it's the current experimental line, so re-check it after a pack update.