EasyUse/Adaptergenerated

Easy Apply IPAdapter(ipadapterApply)

The Easy Apply IPAdapter node (ipadapterApply) applies an IP-Adapter to a diffusion model in one step.

ipadapter Apply

model
image
attn_mask
optional_ipadapter
model
images
masks
ipadapter
preset
COMBO
lora_strength
0.60
provider
CUDA
weight
1.00
weight_faceidv2
1.00
start_at
0.000
end_at
1.000
cache_mode
all
use_tiled
Easy Use

Description

The Easy Apply IPAdapter node (ipadapterApply) applies an IP-Adapter to a diffusion model in one step. It wraps the usual IPAdapter pipeline — preset selection, optional LoRA strength, provider, scheduling, caching, and optional attention mask — and returns a modified model that can be fed directly into a sampler.

This node belongs to the EasyUse/Adapter category.

Inputs

NameTypeRequiredDefaultDescription
modelMODELYes—The diffusion model to apply the IPAdapter to.
imageIMAGEYes—Reference image used by the adapter. For FaceID presets, this should contain a face.
presetCOMBOYes—Selects the IPAdapter preset to use. Available choices depend on your installed presets/models.
lora_strengthFLOATYes0.6Strength for the LoRA used by the selected preset. Presets without a LoRA ignore this.
providerCOMBOYesCUDAExecution provider for the adapter components. Options: CPU, CUDA, ROCM, DirectML, OpenVINO, CoreML.
weightFLOATYes1.0Overall IPAdapter weight. Range: -1 to 3, step 0.05. Set to 0 to disable the adapter’s influence.
weight_faceidv2FLOATYes1.0Separate weight used when the selected preset is FaceID v2. Range: -1 to 5, step 0.05. Ignored by non-FaceID v2 presets.
start_atFLOATYes0.0Step fraction at which the IPAdapter starts being applied. Range: 0 to 1, step 0.001.
end_atFLOATYes1.0Step fraction at which the IPAdapter stops being applied. Range: 0 to 1, step 0.001.
cache_modeCOMBOYesallControls which components are cached between calls. Options: insightface only, clip_vision only, ipadapter only, all, none.
use_tiledBOOLEANYesfalseEnables tiled IPAdapter application to reduce memory usage. Useful for large images, though results may differ slightly.
attn_maskMASKNo—Optional mask to constrain IPAdapter attention.
optional_ipadapterIPADAPTERNo—An existing IPAdapter object to reuse instead of building one from the preset.

Outputs

TypeDescription
MODELThe input model with the IPAdapter applied. Use this in a sampler to generate with the adapter active.
IMAGEThe original input reference image, passed through unchanged.
MASKThe attention mask provided to attn_mask, passed through unchanged, or None if no mask was supplied.
IPADAPTERThe IPAdapter object created or used by this node, available for reuse in other adapter nodes.

Usage Notes

  • With the default scheduling (start_at=0.0, end_at=1.0), the IPAdapter is active across the full denoising range. Lowering end_at applies it only during the early steps, while raising start_at applies it only later.
  • weight is the main strength control. weight_faceidv2 is an additional, separate strength value for FaceID v2 presets.
  • lora_strength controls the LoRA loaded by the selected preset. FaceID-style presets commonly include such a LoRA, but the parameter has no effect for presets that do not include one.
  • cache_mode is useful for batch runs. all caches every reusable component, while none disables caching and frees resources sooner.
  • use_tiled can help avoid out-of-memory errors with high-resolution images. Because tiling can alter attention behavior, compare results if output quality matters.
  • The IMAGE and MASK outputs are passthrough values provided for chaining; they are not generated images or masks.
  • If you supply optional_ipadapter, the node can reuse an already constructed IPAdapter rather than building a new one.

Comments

Sign in with GitHub to join the discussion.

Loading comments…