ComfyUI Node

Image Blend Advance V3

One Photoshop-style blend node, none of the 60-node package around it

By thezveroboy·Created 2 months ago·Updated 2 months ago· 0
Image Blend Advance V3
  • layer_image
  • background_image
  • layer_mask
  • image
  • mask
invert_masktrue
blend_mode
opacity100
x_percent50.00
y_percent50.00
mirror
scale1.00
aspect_ratio1.00
rotate0.00
transform_method
anti_aliasing0

You know that feeling when a workflow needs one node out of a pack of sixty, and ComfyUI drags the other fifty-nine in with it - along with their dependencies and their startup time? That's the entire reason this pack exists. ImageBlendAdvanceV3 is a standalone fork of a single node from chflame163/ComfyUI_LayerStyle, stripped of every external dependency so it ships on its own. No requirements.txt, no models, no hidden installs. The author's whole pitch is that you should be able to grab the one useful node from a bloated package without adopting the package - and honestly, it works.

The node itself is a Photoshop-style layer compositor: it takes a layer image, transforms it (position, scale, rotate, mirror), blends it onto a background using any of 29 blend modes at a set opacity, and masks it. This is the tool for photobashing, pasting a cut-out subject onto a new scene, laying a texture or watermark over a render, or dropping a face back onto a body after an inpaint went sideways. It pairs naturally with the background-removal and inpainting nodes from the wider ecosystem - cut the subject out on one side, blend it back in here.

How it works

Mechanically it's old-school compositing, the kind Photoshop has done for thirty years, wrapped as one node. The layer gets resized, flipped if you asked for mirror, and rotated with canvas expansion (the rotate knob is in degrees, and yes, it's a huge range - don't feed it 999999). Then x_percent and y_percent place it on the canvas as percentages: 50/50 is dead center, 0/0 puts the layer's top-left corner at the canvas top-left. scale is a multiplier on the layer's size, aspect_ratio stretches height relative to width (1.0 keeps proportions), and transform_method picks the resampling filter - lanczos is the default and the one you want for clean results.

The blend is where it earns the "Advance" in the name. Twenty-nine modes, from normal through multiply, screen, color dodge, linear dodge (add), overlay, and the HSV-family modes like hue and saturation. opacity is your 0–100 mix amount, so even at normal you can fade the layer in like an opacity slider in a real editor.

Two outputs: image (the composited result) and mask (the layer's effective mask, built from your input or the layer's alpha channel). That second one is genuinely useful - wire the mask out to feed a detail-enhance or regen node that should only touch the pasted region, not the whole canvas.

The inputs that will bite you

The schema is small, so here's what a beginner actually needs to set:

  • layer_image - the thing you're pasting in. If it's an RGBA image (like something from a background-removal node), its alpha channel becomes the mask automatically.
  • background_image (optional) - the canvas underneath. Leave it empty and you get a transparent canvas the same size as the layer, which is handy when you want a standalone blended result.
  • layer_mask (optional) - a hard mask if you want to control where the layer shows instead of trusting its alpha.
  • invert_mask - the trap. It defaults to true, and it only applies when you actually wire in a layer_mask. If your mask is "keep this region" but the node is silently inverting it, this is the checkbox to blame.

One quirk worth knowing: if your layer_mask's dimensions don't match the layer image, the node throws the mask away and logs a warning rather than failing. You get a fully opaque layer instead - usually not what you wanted, so watch the console.

Install

ComfyUI Manager is the easy path - search "ComfyUI-ImageBlendAdvanceZveroboy" and install. Or from the terminal:

cd ComfyUI/custom_nodes
git clone https://github.com/thezveroboy/ComfyUI-ImageBlendAdvanceZveroboy

Then restart ComfyUI. There's nothing else to install - no requirements.txt in the repo, no model downloads, just torch/PIL/numpy, all of which ComfyUI already ships. In the node menu it shows up as Image Blend Advance V3 under LayerUtility.

The honest verdict

If you already have LayerStyle installed, this pack is redundant - you've got the node. But if you're the kind of setup where startup time and dependency conflicts are the enemy, a single-purpose pack like this is the right trade: it's the node, plus a philosophy you can steal for your own AI-bundled custom nodes. The main things that could trip you up are the inverted-mask default and the mask-size mismatch silently blanking your mask - both are a one-click fix once you know they exist.

CategoryLayerUtility

Inputs (14)

NameTypeDefaultDescription
layer_imageIMAGE
invert_maskBOOLEANtrue
blend_modeCOMBO29 options: normal, dissolve, darken, multiply, color burn, linear burn, +23
opacityINT1000–100
x_percentFLOAT50.00-999–999
y_percentFLOAT50.00-999–999
mirrorCOMBO3 options: None, horizontal, vertical
scaleFLOAT1.000.01–100
aspect_ratioFLOAT1.000.01–100
rotateFLOAT0.00-999999–999999
transform_methodCOMBO6 options: lanczos, bicubic, hamming, bilinear, box, nearest
anti_aliasingINT00–16
background_imageoptIMAGE
layer_maskoptMASK

Outputs (2)

NameTypeDescription
imageIMAGE
maskMASK