Apply Regional IPAdapters (Inspire)
The node that actually applies your regional IP-Adapters to the model
- ipadapter_pipe
- regional_ipadapter1
- MODEL
This is the node that finishes the regional IP-Adapter job. All the Regional IPAdapter … nodes (by mask, by color mask, encoded, and so on) each produce a REGIONAL_IPADAPTER handle describing "this reference, in this region." They don't touch your model on their own. This node takes those handles plus an IP-Adapter pipe and bakes them into a MODEL you then sample with. If you've built up per-region adapters and wondered where they actually get applied - here.
In the wider workflow, regional IP-Adapter is how you give different parts of one image different visual references: left character styled from reference A, right from reference B, background from C. You define each region upstream, then collect them here and apply them all at once so the sampler sees a single model conditioned region by region.
How it works
IP-Adapter injects a reference image's features through a dedicated cross-attention path, leaving the text path alone. The regional nodes scope that injection to a masked area. This node walks the regional adapters you hand it and patches all of them onto the base model carried in the IP-Adapter pipe, producing a modified MODEL. That model output is what you feed your KSampler (or custom sampler) - from there it samples normally, but each region pulls toward its assigned reference.
The ipadapter_pipe bundles the pieces IP-Adapter needs (the IP-Adapter model, CLIP Vision, and the base model). Inspire's ToIPAdapterPipe node assembles that bundle; this node consumes it.
The inputs and outputs that matter
ipadapter_pipe(IPADAPTER_PIPE) - the bundled IP-Adapter model + CLIP Vision + base model. Build it withToIPAdapterPipe (Inspire).regional_ipadapter1(REGIONAL_IPADAPTER) - your first region's adapter. The node accepts more regional adapters as you add them; wire one per region you defined upstream.
The single output is MODEL - the base model with all your regional IP-Adapters applied. Feed it straight into your sampler.
How to install it
ComfyUI Manager: search ComfyUI Inspire Pack, install, restart. Manual:
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Inspire-Pack
then restart. It's a Dr.Lt.Data pack (ComfyUI-Manager, Impact Pack).
This whole regional-IPAdapter family depends on cubiq's ComfyUI_IPAdapter_plus - the Inspire README says as much. Install it via Manager too, or the nodes won't function. And per the knowledge base, IP-Adapter is an SD1.5 / SDXL tool; there's no Flux-era equivalent in this form.
Common issues & troubleshooting
"IPAdapterPlus is not installed" / the regional nodes are broken upstream. cubiq's IPAdapter Plus pack is missing. Install it via Manager and restart. This apply node can't do anything if the regional adapters feeding it never initialized.
No visible regional effect. Two usual causes. First, check the ipadapter_pipe actually carries the right IP-Adapter model and CLIP Vision for your base architecture. Second, the per-region weight (set on the regional nodes) may be too low, or the start_at/end_at window too short - the effect is applied there, not here.
Only one region shows up. Make sure every region's REGIONAL_IPADAPTER is wired into this node, not just the first. Each region you defined has to reach this node to be applied.
Wrong architecture. The IP-Adapter model, CLIP Vision, and base model in the pipe must all match one architecture (all SD1.5, or all SDXL). Mismatches either error or quietly produce garbage. Keep the whole chain on one base.
It won't sample. The output is a MODEL - feed it into the sampler's model input, replacing the original model there. If you sample with the un-applied model, none of this took effect.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| ipadapter_pipe | IPADAPTER_PIPE | — | |
| regional_ipadapter1 | REGIONAL_IPADAPTER | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |