✂️ IC Lora Layout Crop

This is the back-half of a two-node pair for IC-LoRA (In-Context LoRA) workflows, and it makes almost no sense on its own - so here's the context first. IC-LoRA is a Flux-era trick where, instead of conditioning on a reference the normal way, you stitch several images into one multi-panel canvas and let the model treat the whole grid as a single image. Fill one panel with a reference and leave the other blank, and the model generates the empty panel in the style of the filled one - same character, same outfit, consistent look - because to the model they're literally the same picture.

The catch: your result comes out as one wide multi-panel image, and you need the individual panels back. That's this node. The companion IC Lora Layout node builds the combined canvas and hands out a CROP descriptor recording where each panel sits; IC Lora Layout Crop takes that descriptor and slices the generated image back into the piece you actually want.

How it works

The layout node knows the geometry - how many panels, where each one is, how they were padded. Rather than make you re-measure that by hand, it emits a CROP object carrying those boundaries. You pass the finished (generated) image plus that CROP into this node, and it extracts the corresponding region as a clean, standalone image. It's a deterministic crop, not a smart detector - the coordinates come straight from the layout step, so the split is exact.

The inputs and outputs that matter

Two inputs, one output, and both inputs are required:

• image - the multi-panel image to crop, normally the KSampler output after the IC-LoRA pass.
• crop - the CROP descriptor from the IC Lora Layout node. This is the piece that carries the panel geometry; you can't hand-type it, it has to come from the layout node.

Output is a single image - the extracted panel.

How to install it

ComfyUI Manager: search SDVN_Comfy_node, install, restart. Or clone:

cd ComfyUI/custom_nodes
git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node

then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root, and restart. Both IC-LoRA nodes are under SDVN → Image. You'll also need an actual IC-LoRA (a Flux In-Context LoRA) for the generation step - this node only handles the crop, not the model.

Common issues & troubleshooting

No crop input available / type mismatch. The crop slot only accepts the CROP output of the IC Lora Layout node. If you're trying to feed it a mask, a number, or a plain image, it won't connect - the two nodes are designed as a set and the descriptor is the glue between them.

The crop is offset or grabs the wrong panel. That means the layout and crop steps disagree about geometry, usually because the image was resized or re-padded between them. Keep the generated image at the same dimensions the layout produced, and don't insert a resize in between.

"Do I even need this?" Only if you're doing IC-LoRA panel workflows. For ordinary reference conditioning, an editing model or IP-Adapter is the simpler road; IC-LoRA is a specific technique for tight consistency across panels, and this node is purely its cleanup step.