Regional Conditioning Simple (Inspire)

This is the lightweight way to do regional prompting in ComfyUI. It takes a clip, a mask, and a prompt, encodes the prompt, restricts it to the masked area, and hands you back a CONDITIONING you can combine with others and drop into an ordinary KSampler. No separate regional sampler, no basic_pipe, no per-region sampling pass - just masked conditioning you assemble yourself.

Under the hood it's a convenience wrapper around ComfyUI's built-in Conditioning (Set Mask). That node already lets you constrain a prompt to a region; this one just folds the encode-and-mask into a single step so your graph reads cleaner. Reach for it when you want two or three subjects to keep their own attributes and you're happy driving a normal sampler - the classic "girl on the left in red, girl on the right in blue" without them trading colours. It's the cheaper cousin of the Regional Prompt nodes, which run a full sampling pass per region through the Impact Pack.

How it works

You build one of these per region. Each encodes its prompt against the clip, then uses the mask to limit where that conditioning applies. You wire several together with a Conditioning (Combine) and pass the result to a standard KSampler as its positive conditioning. The sampler blends the regions during generation.

The inputs and outputs that matter

• clip - the CLIP to encode this region's prompt with.
• mask - the area this prompt covers.
• prompt - the region's text.
• strength - how strongly this region's conditioning pushes (default 1).
• set_cond_area - default applies over the whole latent weighted by the mask; mask bounds tightens the conditioning to the mask's bounding box, which can sharpen where the effect lands.

Output: a single CONDITIONING - combine it with your other regions and feed a KSampler.

Installing it

ComfyUI Manager: search "Inspire Pack", install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack

Then restart. There's a requirements.txt Manager installs for you. Nice thing about this particular node: because it wraps a core ComfyUI feature, it doesn't drag in a sibling pack the way the regional sampler and IPAdapter nodes do.

Where people get tripped up

Masked conditioning is the loosest form of regional control, so expect some bleed at the edges - text-token masking sets soft bounds, and attributes can leak across region borders. That's a known tradeoff of the technique, not a bug in the node; if the bounds have to be tight, the heavier regional-sampler route (Regional Prompt Simple into Impact's sampler) holds them better at the cost of setup and render time. Also remember to actually combine your regions and set a base/background conditioning - a single masked region with nothing covering the rest of the canvas leaves that area under-conditioned and mushy.