Regional Prompt Simple (Inspire)

Regional prompting is how you stop two characters in one image from swapping hair colour and clothes. Prompt adherence is global - describe a red-haired woman and a blue-haired woman in one prompt and the model reliably smears them together - and no amount of prompt wording fixes it. You split the canvas into regions and give each its own prompt. This node is the ergonomic front door to doing that with the Impact Pack's regional sampler: hand it a mask and a basic_pipe and it produces a REGIONAL_PROMPTS entry with almost no wiring.

It exists because the raw way of setting this up in Impact is fiddly. Regional Prompt Simple collapses "here's the area, here's the prompt, here's how to sample it" into one node. You chain a few of them - one per region - into the Impact Pack's Regional Sampler, and each region gets sampled with its own prompt inside its own mask.

How it works

The regional sampler doesn't just mask conditioning; it actually samples each masked region with its own settings and composites the result. So this node bundles everything that region needs: the basic_pipe carries the model/clip/vae and base prompts, the mask says where, and the node's own wildcard_prompt plus sampler settings say what and how. Its output is a REGIONAL_PROMPTS object - the format Impact's Regional Sampler consumes.

The inputs and outputs that matter

• basic_pipe - the bundled model/clip/vae + base conditioning (build it with Make Basic Pipe (Inspire) or Impact's own pipe nodes).
• mask - the region this prompt applies to.
• wildcard_prompt - the text for this region; supports wildcard syntax since it runs through the pipe's encoder.
• cfg / sampler_name / scheduler - the sampling settings for this region specifically.
• sigma_factor - scales how strongly this region is denoised relative to the base; a subtle but useful dial for making a region blend in or stand out.

Output: REGIONAL_PROMPTS - feed it (chained with other regions) into the Impact Pack Regional Sampler.

Installing it

ComfyUI Manager: search "Inspire Pack", install, restart. Manually:

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

Restart after. The pack has a requirements.txt (Manager handles it automatically).

Where people get tripped up

This node is a helper for the Impact Pack, not a standalone regional system - REGIONAL_PROMPTS goes nowhere without Impact's Regional Sampler downstream, so install the Impact Pack too (same author, so grab it while you're there). The common quality complaint isn't the node, it's the technique: hard region masks held through the whole sample produce a collaged look with visible seams. The fix the community settled on is to let composition form and then blend, which is partly why the sigma_factor and variation controls exist. And if you'd rather not run a separate sampling pass per region at all, the lighter Regional Conditioning Simple route just masks the conditioning and feeds one normal KSampler - less control, far less setup.