AIHub Action New Layer

AIHub Action New Layer is the action node that understands the client has a canvas with layers. Instead of just handing an image back like AIHub Action New Image does, it tells the image editor to create a layer - positioned at an X/Y coordinate, inserted before or after a specific existing layer, with a name that shows up in the layer list. This is the node that makes "AI as an inpainting or compositing tool inside your editor" workflows work: you generate a patch in ComfyUI and it arrives in the editor already aligned with everything else.

Pack context, once: otavanopisto's ComfyUI-aihub-workflow-exposer is the bridge that lets external apps run ComfyUI as their engine. Expose nodes pull data in from the client (including the current layer, its position, and its mask - that's where your pos_x/pos_y come from), and action nodes like this one push results back over the pack's own websocket (port 8111). The whole design revolves around the layer metaphor of a real image editor.

How it works

Same encode pipeline as New Image - tensor to PNG, mask folded into the alpha channel - but the websocket event is NEW_LAYER, and the envelope carries pos_x, pos_y, reference_layer_id, and reference_layer_action. The client receives both the image data and the placement instruction, then builds the layer in its own stack. The __first__ and __last__ special values for reference_layer_id let you pin a layer to the top or bottom of the stack without knowing the client's internal ids.

The inputs that matter

• image (required) - what becomes the layer.
• pos_x, pos_y - canvas coordinates for the layer. In a typical workflow these come from an AIHub Expose Image node's POS_X/POS_Y outputs, so the generated result lands exactly where the source layer was.
• reference_layer_id - the layer to position against; __first__ and __last__ are the handy shortcuts.
• reference_layer_action - NEW_AFTER, NEW_BEFORE, or REPLACE. Replace swaps out the referenced layer's content, which is the natural choice for "regenerate this layer in place."
• action - the filesystem-level APPEND/REPLACE for the stored file. The tooltip is worth quoting: since layers are meant to be integrated into the project, REPLACE is often right - the client may delete the old file later anyway.
• mask (optional) - alpha for the layer, e.g. a soft-edged inpaint region.
• name - the layer's label in the editor's layer panel.

It echoes IMAGE and MASK back; the real work is the side effect.

Installing it

ComfyUI Manager (search "ComfyUI-aihub-workflow-exposer"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/otavanopisto/ComfyUI-aihub-workflow-exposer
# restart ComfyUI

No requirements, no downloads - it uses ComfyUI's own Pillow/torch stack.

Where people get burned

Position mismatches are the classic annoyance: if the coordinates come from a source image exposed in a different mode than the one your editor supports (see the many type options on Expose Image), your layer lands offset from where you expect. Trace where pos_x/pos_y originate before blaming the node. And the pack-wide rule applies - no aihub client connected means the layer is created nowhere, so a "nothing happened" run is a connection problem, not a node problem.