omost Setting

Omost, lllyasviel's regional-composition technique (the same author behind IC-Light), works by breaking a scene into separate areas - a color, a rough position, an offset, and a size for each - instead of one flat prompt. Normally you'd have an LLM improvise that breakdown. omost Setting is the manual alternative: four dropdowns, no model call, no randomness. You pick exactly what one region should be.

comfyui_LLM_party credits Omost directly in its own README as one of the projects it borrows from, and this node sits alongside the pack's omost json2py (which converts an LLM-generated JSON layout into Omost's Python format). Where that node handles the automated path, omost Setting is for when you want deterministic, hand-picked control over a single region instead of trusting an LLM to guess your intent.

How it works

Four enum dropdowns, each describing one dimension of a region, get combined into a single descriptive string. There's no free text here - every field is a fixed choice from a preset list, which is the whole point: you can't typo a region description or accidentally produce something Omost's downstream renderer can't parse, because every option is a known-good phrase.

The inputs and outputs that matter

• color - a CSS-style color name (149 choices, from aliceblue through the full named-color list). This sets the region's dominant color.
• locations - where in the frame the region sits: in the center, on the left, on the right, on the top, on the bottom, and the four diagonal corners.
• offsets - a fine nudge from that base location: no offset, or slightly to the upper/lower/left/right and the diagonal combinations.
• areas - the region's rough size and shape: small/medium/large, crossed with square/vertical/horizontal.

Output: string (STRING) - the assembled region description, ready to be fed into whatever's building your Omost layout, alongside any other regions you're constructing the same way.

Where it fits

Building a multi-region scene by hand means placing one of these nodes per region and combining the resulting strings (this pack's String Combine nodes are the natural pairing) into the full layout Omost expects. It's more tedious than letting an LLM improvise the whole thing in one shot via omost json2py, but it buys you certainty: no risk of the model inventing a nonsensical position or a color name that doesn't parse. Reach for this when you know exactly what you want and don't want to fight an LLM's interpretation of your instructions; reach for the LLM-driven path when you want variety or don't want to place every region yourself.

How to install it

Search comfyui_LLM_party in ComfyUI Manager and install, then restart. Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/heshengtao/comfyui_LLM_party.git

Run pip install -r requirements.txt from the pack's folder with ComfyUI's own Python, then restart. This node itself needs nothing beyond the base install - it's pure dropdown-to-string assembly. The actual Omost rendering step downstream is a separate concern from this pack and has its own requirements.

Common issues & troubleshooting

The composed layout doesn't look like what you expected. This node only produces a text description of a region - it doesn't render anything itself, so if the final image looks off, the issue is more likely in how the downstream Omost renderer interprets combined region strings, or in how many regions you've stacked, than in this node's output.

You want more precision than the dropdowns give you. These fields are deliberately coarse presets, not continuous values - there's no way to place a region at an exact pixel offset here. If you need that level of control, you're past what this node is designed for and should look at a more direct regional-conditioning approach instead.