Shima Panel Latent Maker

Shima Panel Latent Maker is the size-and-shape node in Shima's Panel workflow. It creates the empty latent tensor the sampler starts from, and it does it with aspect-ratio presets instead of bare pixel numbers. On the canvas it's the usual compact PCB chassis with one latentmaker.bndl output; double-click for the controls: aspect_ratio, orientation (portrait/landscape), a scale multiplier, width/height, and batch_size.

If you've ever stood at an Empty Latent Image wondering what dimensions to type for a portrait image on SDXL, this is the node that fixes that. Pick an aspect ratio, flip orientation, adjust scale - the node does the math and gives you a properly rounded, model-appropriate latent. The model_type setting (SDXL, SD 1.5, and the rest) keeps the defaults sane per architecture, and the whole thing reports both the raw latent and a bundle that carries the dimensions with it.

Why the bundle matters: latentmaker.bndl rides the width, height, and seed alongside the latent. When the Panel Sampler consumes it, it can read the intended size without extra wires - that's the same "one wire carries a context" pattern that keeps Shima graphs tidy.

How it works

It's the panelized version of the Latent Maker node. Given a model_type, an aspect_ratio, an orientation, a scale, and explicit width/height fallbacks, it computes final pixel dimensions and allocates an empty latent tensor at the right resolution for the chosen architecture, with batch_size copies for batch generation. The seed handling comes via s33d and s33d_mode (fixed/randomize-style modes), and like all panels it can defer to Shima's shared Commons node when use_commonparams is on - the green/red circle in the topbar.

The inputs and outputs that matter

Required: s33d, s33d_mode, model_type, aspect_ratio, orientation, width, height, scale, batch_size. Optional: shima.commonparams, use_commonparams, allow_external_linking, show_used_values, and panelinputs.bndl (external override from a Panel BNDLer). Outputs are latent (LATENT) for non-BNDL consumers and latentmaker.bndl for the Panel Sampler.

How to install it

In the Shima pack. ComfyUI Manager → search "Shima", or:

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf.git Shima

restart. Core deps install automatically; no model downloads - this node only creates tensors.

Common issues & troubleshooting

Generated image is a different size than you set. Check whether use_commonparams is on - with it on, the Commons node's width/height/seed override the panel's local values, which is exactly what "Common Params" is supposed to do. If you want local control, toggle the sync circle off.

Output is off-model (e.g. non-multiple-of-8 sizes). The model_type should match the checkpoint feeding the sampler. The node rounds to architecture-appropriate buckets; if you force a weird width/height via the override bundle you can bypass that rounding.