Send to Editor Tab

This is the other half of Lerc's Canvas Tab pack, and it's the return trip. Where Edit In Another Tab reads a painted canvas and mask out of the editor, Send to Editor Tab pushes your generated images back into it. On its own it does nothing you can see - it's an output node with no outputs - but paired with the editor node it's what closes the loop and makes the whole thing feel like a paint program instead of a graph.

How it works

It takes one optional input, images (an IMAGE), converts each incoming tensor to a PNG data URL, and ships it to the editor page through the same browser channel the editor node uses. Where the image lands depends on the Incoming Images dropdown in the editor: Replace Targeted Layer (the default), Add layer, Create new Image, or Ignore. Right-click a layer in the editor to set it as the target for replacement.

The detail that makes it tick is IS_CHANGED: the node fakes a change on every execution, so it always runs even when its upstream inputs are byte-identical. That's what keeps a live-drawing loop flowing. Contrast that with Edit In Another Tab, which only re-runs when its own inputs change - this node is the workaround that keeps fresh generations streaming into the editor.

You can also drag an image file straight onto the node; it's the same code path, handed directly to the editor.

The loop it enables

With a turbo model the pattern is: scribble in the editor → generate → this node drops the result back in as a layer → Ctrl-click the Duplicate Layer button to copy it as a fresh base → draw on top at a lower denoise → repeat. The repo's workflow folder has an SDXL Turbo example built around exactly this, including a sharpening node you can flip on when repeated iteration starts getting blurry.

Installing

Same as the pack - no separate install. ComfyUI Manager (search "Canvas Tab"), or:

cd ComfyUI/custom_nodes
git clone https://github.com/Lerc/canvas_tab

Then restart ComfyUI; the node lives in the image submenu. No extra dependencies, no model downloads.

Gotchas

• No output, don't panic. It's meant to be a sink. If the editor window is closed when a generation lands, the node opens it for you (or retries half a second later if the channel isn't ready).
• Title matching works here too. If you have several Edit In Another Tab nodes, naming an editor image to match a node's title routes results to the right one.
• The pack is quiet these days. Development slowed in 2024 and community chatter dried up by 2026 as instruction-editing models ate most of masked inpainting's audience. It still works fine - just don't expect updates, and read the Canvas Tab article for the honest appraisal of when hand-painted masks are still the right call.