EasyUse/Adaptergenerated

Easy Apply InstantID (Advanced)(instantIDApplyAdvanced)

Easy Apply InstantID (Advanced) node from ComfyUI-Easy-Use.

instant ID Apply Advanced

pipe
image
image_kps
mask
control_net
positive
negative
pipe
model
positive
negative
instantid_file
COMBO
insightface
COMBO
control_net_name
COMBO
cn_strength
1.00
cn_soft_weights
1.000
weight
0.80
start_at
0.000
end_at
1.000
noise
0.35
Easy Use

Applies InstantID identity using the advanced Easy-Use pipeline. This node loads an InstantID model, detects the reference face with InsightFace, wires an optional ControlNet for structural guidance, and injects the identity into the model and conditioning in one step. It works from a PIPE_LINE and can also return the modified MODEL and positive/negative CONDITIONING separately.

The node is part of the EasyUse/Adapter category and is intended for workflows that need finer control over InstantID strength, ControlNet influence, and scheduling than the basic Easy Apply InstantID node.

Inputs

InputTypeRequiredDefaultDescription
pipePIPE_LINE✅—The pipeline state containing the model, conditioning, and other workflow data.
imageIMAGE✅—Reference image containing the face whose identity will be applied.
instantid_fileCOMBO✅—InstantID model file to load.
insightfaceCOMBO✅—Provider used for face detection/analysis: CPU, CUDA, or ROCM.
control_net_nameCOMBO✅—ControlNet model name used for structural pose/face guidance.
cn_strengthFLOAT✅1Strength of the ControlNet conditioning. Range: 0 to 10, step 0.01.
cn_soft_weightsFLOAT✅1Softness of ControlNet weights. Range: 0 to 1, step 0.001.
weightFLOAT✅0.8Overall strength of the InstantID identity injection. Range: 0 to 5, step 0.01.
start_atFLOAT✅0Scheduling start point for InstantID/ControlNet influence during sampling. Range: 0 to 1, step 0.001.
end_atFLOAT✅1Scheduling end point for InstantID/ControlNet influence during sampling. Range: 0 to 1, step 0.001.
noiseFLOAT✅0.35Amount of noise mixed into the identity embedding. Lower values usually stay closer to the reference face. Range: 0 to 1, step 0.05.
image_kpsIMAGE❌—Optional keypoint/pose image for ControlNet structural guidance.
maskMASK❌—Optional mask to restrict where the InstantID/ControlNet conditioning is applied.
control_netCONTROL_NET❌—Optional pre-loaded ControlNet, useful when you already have one configured instead of loading by control_net_name.
positiveCONDITIONING❌—Optional positive conditioning. If omitted, the positive conditioning from pipe is used.
negativeCONDITIONING❌—Optional negative conditioning. If omitted, the negative conditioning from pipe is used.

Outputs

OutputTypeDescription
PIPE_LINEPIPE_LINEUpdated pipeline with the InstantID-applied model and conditioning.
MODELMODELThe modified model after InstantID and ControlNet injection.
CONDITIONINGCONDITIONINGPositive conditioning adjusted for InstantID.
CONDITIONINGCONDITIONINGNegative conditioning adjusted for InstantID.

Usage Notes

  • Use image for the face whose identity you want to preserve. The instantid_file determines which InstantID model is used for the identity embedding.
  • insightface should match your runtime. CUDA is usually fastest on NVIDIA GPUs, ROCM on AMD GPUs, and CPU works everywhere but is slower.
  • cn_strength and cn_soft_weights control the ControlNet side of InstantID. The ControlNet is typically used to keep pose or facial structure consistent while the identity is applied.
  • weight is the main dial for how strongly the reference identity affects the generated result.
  • start_at and end_at let you schedule when InstantID starts and stops influencing the denoising process. Restricting this range can help preserve composition early or avoid over-constraining later steps.
  • noise affects the identity embedding variation. Lower values generally produce a closer match to the reference face; higher values can allow more expression or variation.
  • When positive and negative are provided, they are used as the starting conditioning for InstantID. When omitted, the node uses the conditioning already stored in pipe, which makes it convenient for chaining from an Easy-Use sampler/loader pipeline.
  • Returning MODEL and CONDITIONING outputs lets you use this node without keeping the whole PIPE_LINE, or to feed the modified values into other custom nodes.
  • If you provide an external control_net, it takes precedence over loading from control_net_name. This is useful when you want to share a single loaded ControlNet across multiple nodes.

Comments

Sign in with GitHub to join the discussion.

Loading comments…