Nodes/ComfyUI-StyleTransferPlus/AesFA Styles Blending
ComfyUI Node

AesFA Styles Blending

Borrow the palette from one painting and the brushstrokes from another

By FuouM·Created 2 years ago·Updated about a year ago· 15
AesFA Styles Blending
  • src_img
  • style_hi
  • style_lo
  • out_img
do_cropfalse
size512

This is the party trick of the ComfyUI-StyleTransferPlus pack. AesFA Styles Blending runs the exact same AesFA model as the plain AESFA node, but instead of one style image it takes two - and it deliberately pulls different things from each. The pack author sums it up better than I can:

The style-transferred outputs finely keep the color information from the low-frequency image and change the texture information based on the high-frequency image.

So: one painting gives you the mood and palette, the other gives you the brushwork. If you've ever mixed a style LoRA and wished you could decouple "colors" from "texture" with a slider, this is the closest thing in a node graph - except here there's no diffusion, no prompt, and no slider. Just two images in.

How it works

Remember AesFA splits a style into a low-frequency band (color, tone) and a high-frequency band (texture)? This node wires that decomposition into the input ports directly. style_lo feeds the low-frequency path - that's where your colors come from. style_hi feeds the high-frequency path - that's your texture and stroke pattern. The content features get convolved with aesthetic feature-aware kernels predicted from both, and the two influences are recombined in the output.

Same quirks as the single-image version apply:

  • Square-only. Everything is resized to size×size; do_crop makes it resize-then-center-crop instead.
  • The known AesFA vertical line artifacts can show up here too, since it's the same network under the hood.

Inputs and outputs

  • src_img - the content you're restyling.
  • style_lo - the low-frequency style: your color/tone source.
  • style_hi - the high-frequency style: your texture/brush source.
  • size (default 512) and do_crop - same as the plain AesFA node.

Output is out_img, an IMAGE, straight into Preview/Save. To steal a specific palette, put the colorist painting in style_lo; to steal a texture, put the textured one in style_hi. Swapping the two flips the result dramatically, which makes it a genuinely fun node to play with.

Install and setup

Same pack, same steps as AESFA: install via ComfyUI Manager (search ComfyUI-StyleTransferPlus) or

cd ComfyUI/custom_nodes
git clone https://github.com/FuouM/ComfyUI-StyleTransferPlus

restart, then drop the single required model file inside the pack folder:

ComfyUI/custom_nodes/ComfyUI-StyleTransferPlus/models/aesfa/main.pth

That one main.pth serves both AesFA nodes - if you already downloaded it for AESFA, this node just works.

Common issues

  • Model not found - the models/aesfa/main.pth path inside the custom node folder is the only thing this node needs; if it's missing you'll see a load error on first run.
  • Output looks like only one style came through - check which image is in which port. Colors live in style_lo, texture in style_hi. Get them backwards and it still runs, but you'll wonder where your palette went.
  • Square output / vertical artifacts - inherited from AesFA; a smaller size is your main lever.

It's a niche feature of a niche pack, honestly - you won't find a wall of tutorials about it. But if you want fine control over which part of a style you keep, there isn't another node in ComfyUI doing quite this trick.

CategoryStyleTransferPlus

Inputs (5)

NameTypeDefaultDescription
src_imgIMAGE
style_hiIMAGE
style_loIMAGE
do_cropBOOLEANfalse
sizeINT512

Outputs (1)

NameTypeDescription
out_imgIMAGE