Forbidden Vision 🧩 Face Edit Merge
The node that puts your edited face back where it belongs
- original_image
- edited_face
- face_info
- final_image
- compositing_mask
Face Edit Merge is the second half of the Forbidden Vision "edit the crop yourself" workflow, and it solves the problem that kills most hand-rolled face edits: getting the result back onto the original without a visible seam. If Face Edit Prep is the node that grabs the face, this is the node that stitches it home.
What it does
You give it three things:
original_image- the untouched full image.edited_face- your edited face, at the same resolution as thecropped_facethat Prep handed you.face_info- the geometry packet from the Prep node. This is what makes the whole thing work: it encodes where the crop sat in the original, the rotation that was applied during prep (so the merge can undo it), and the mask shape. Without it, "put this face back" would be guesswork.
It then composites the edited face over the original using the stored mask, feathers the transition with blend_softness (default 8), and - if enable_color_correction is on (it is by default) - shifts the edited face's tone and lighting to match the surrounding original, with color_correction_strength (default 1.0, up to 2.0) controlling how hard it pushes. This is the step that stops the classic tell of a pasted face: right shape, wrong lighting.
Outputs
Two: final_image (the composite) and compositing_mask, which is the mask it actually used to blend - handy for previewing exactly which pixels changed, or for stacking another operation on the same region.
Why you'd reach for it instead of the Fixer
The all-in-one Fixer samples the face with its own built-in inpainting, which is fine when you just want a better face. But there's a whole class of edits the Fixer can't do: putting back a face you touched up in Photoshop, pasting a face generated by a different model or workflow, or compositing a crop you edited with a node the pack doesn't know about. The Prep + Merge pair lets you own the middle step - anything that produces an image at the right resolution works.
The practical trap is resolution: edited_face must match the cropped_face output from Prep, or the merge will warp it. Keep the crop resolution and aspect ratio intact through whatever editor you use, and the blend will do its job. If the seam still shows after that, the dials to reach for are blend_softness (up, to feather wider) and color_correction_strength (up, to force a closer tone match). It's a small, focused node - five inputs, two outputs, no hidden model downloads - and it's the piece that makes the "edit the face your way" workflow actually pleasant instead of a paste-and-pray exercise.
Inputs (6)
| Name | Type | Default | Description |
|---|---|---|---|
| original_image | IMAGE | Original full image. | |
| edited_face | IMAGE | Edited face image at the same resolution as the cropped face from prep node. | |
| face_info | FACE_INFO | Face info output from the prep node. | |
| blend_softness | INT | 80–200 | — |
| enable_color_correction | BOOLEAN | true | — |
| color_correction_strength | FLOAT | 1.000–2 | — |
Outputs (2)
| Name | Type | Description |
|---|---|---|
| final_image | IMAGE | — |
| compositing_mask | MASK | — |