Nodes/ComfyUI Inspire Pack/Regional IPAdapter Mask (Inspire)
ComfyUI Node Runs on cloud

Regional IPAdapter Mask (Inspire)

Apply an image prompt to just one part of the frame

By ltdrdata·Created 3 years ago·Updated 9 months ago· 805
Regional IPAdapter Mask (Inspire)
  • mask
  • image
  • neg_image
  • REGIONAL_IPADAPTER
weight0.70
noise0.50
weight_type
start_at0.000
end_at1.000
unfold_batchfalse
faceid_v2false
weight_v21.00
combine_embeds

IPAdapter lets you prompt with an image instead of text - feed it a reference and the model borrows that face, style, or subject. This node restricts that image prompt to a masked region. So you can put one reference character on the left of the frame and a different one on the right, or apply a style reference to the background only, by pairing each image with a mask. It's regional prompting, but for image conditioning rather than text.

It's a convenience wrapper around cubiq's ComfyUI IPAdapter Plus attention-mask feature. Setting that up by hand is a chore; this node bundles the reference image, the region mask, and the IPAdapter weight settings into one REGIONAL_IPADAPTER object that the downstream IPAdapter apply node consumes. Chain a few and you've got per-region image prompting with minimal wiring.

How it works

You hand it a mask (where) and an image (the reference to apply there), plus the usual IPAdapter dials. It packages those into a REGIONAL_IPADAPTER. You combine one per region and pass the bundle into the IPAdapter application step, which masks each reference's influence to its area during sampling.

The inputs and outputs that matter

  • image - the reference image prompt for this region.
  • mask - the area it applies to.
  • weight (default 0.7) - how strongly the reference pushes; the single most important dial, same as any IPAdapter.
  • start_at / end_at - the fraction of sampling steps over which this reference is active; pulling end_at back lets structure form before the reference takes hold.
  • weight_type - the IPAdapter weighting mode (this list is populated by IPAdapter Plus; see below).

Output: REGIONAL_IPADAPTER - combine with other regions and feed the IPAdapter apply node.

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. Critically, you also need cubiq's ComfyUI IPAdapter Plus installed - this node is a thin wrapper over its masking feature and does nothing without it.

Where people get tripped up

There's a dead giveaway when the dependency is missing: the weight_type dropdown literally reads "IPAdapterPlus is not installed" instead of the usual mode list. If you see that, stop - install cubiq's ComfyUI IPAdapter Plus, restart, and the node comes to life. The Inspire README also warns this node is incompatible with outdated IPAdapter Plus builds (it wants a reasonably recent version), so update both together via Manager. Beyond the dependency, the usual IPAdapter rules apply: you need the matching IPAdapter and CLIP Vision models downloaded, and weight around 0.7 is a sane start - crank it and the reference takes over the whole region, drop it and it barely registers.

CategoryInspirePack/Regional

Inputs (12)

NameTypeDefaultDescription
maskMASK
imageIMAGE
weightFLOAT0.70-1–3
noiseFLOAT0.500–1
weight_typeCOMBO1 options: IPAdapterPlus is not installed
start_atFLOAT0.0000–1
end_atFLOAT1.0000–1
unfold_batchBOOLEANfalse
faceid_v2optBOOLEANfalse
weight_v2optFLOAT1.00-1–3
combine_embedsoptCOMBO5 options: concat, add, subtract, average, norm average
neg_imageoptIMAGE

Outputs (1)

NameTypeDescription
REGIONAL_IPADAPTERREGIONAL_IPADAPTER