Omost Render Canvas Conditioning

Omost Render Canvas Conditioning is the visual debugger of the pack. Feed it OMOST_CANVAS_CONDITIONING and it outputs two things: an IMAGE and a MASK. The image is the "initial latent" - a 90×90 color map where each region is filled with the color the LLM assigned, sitting on top of the global background color from the first canvas entry. The mask output is the per-region binary masks, one per condition, stacked into a single tensor.

What's it actually for? Three things, in order of how often people use them.

First, preview. This is the big one. You can finally see whether the layout matches your intent before burning minutes of sampling - with the LLM path each layout is expensive, so a two-second sanity check saves real time. A layout you never look at will absolutely surprise you.

Second, initialization. The original Omost repo recommends passing the rendered canvas image as a starting point for diffusion rather than pure noise, so the composition roughly follows the color blobs. Encode the IMAGE with a VAE encode and use it as the initial latent; it's a rough scaffold, not a painting, but it steers the early composition steps.

Third, debugging the mask math. When regions overlap and you're fighting seams or regions that refuse to stay put, seeing the raw rectangles makes the problem obvious - usually it's two boxes sharing an edge you didn't intend.

One honest caveat: it renders at 90×90, the coordinate space Omost operates in, so don't expect a usable standalone image. Your sampler upscales it to real resolution, and the colors are just placeholders. Also note the outputs are one whole-canvas image plus a batched mask - not one image per region. If you want per-region masks to, say, apply a ControlNet or IPAdapter inside a single box, the MASK output on the Omost Layout Cond (ComfyUI-Area) node is the more convenient shape. For everything else, this node is the fastest way to know whether the layout you wrote is the layout you meant.