Enchance And Resize Hint Images

Yes, the node is spelled "Enchance" - search for it exactly if you're hunting it in the menu. Typo aside, it does something genuinely useful and unglamorous: it takes an already-made hint image (a depth map, edge map, pose skeleton, whatever) and resizes it to your target generation dimensions, using the same resize strategies the old sd-webui-controlnet extension offered. When your control map and your canvas don't share an aspect ratio, this is the node that reconciles them cleanly instead of letting something stretch or crop badly.

What it's for

Control maps often come in at one shape and your render wants another. Maybe you preprocessed a 4:3 photo but you're generating a 16:9 image. Naively forcing that mismatch distorts the structure your ControlNet is trying to follow, and distorted structure means a distorted result. This node gives you deliberate control over how the hint gets fit: stretch it, crop it, or letterbox it. It's plumbing between a preprocessor and a ControlNet Apply node, most valuable exactly when aspect ratios don't line up.

The "enchance" half is a light touch-up of the hint as it's resized, so the map stays crisp rather than going soft from the resample.

The inputs and output

• hint_image - the control map you've already generated (this node doesn't make hints, it fits them).
• image_gen_width (default 512) and image_gen_height (default 512) - your target render dimensions. Set these to match the resolution you're actually generating at.
• resize_mode - the important choice, with three options:
  • Just Resize - stretch the hint to the target dimensions, ignoring aspect ratio. Fine when the ratios already match; distorts when they don't.
  • Crop and Resize - scale to fill, then crop the overflow. Keeps proportions, loses edges. Good when the subject is centered.
  • Resize and Fill - scale to fit, then pad the gaps (letterbox). Keeps the whole map, adds borders.

The single output is an IMAGE - the resized hint - which continues on into your ControlNet Apply node.

Installing it

ComfyUI Manager: search ComfyUI's ControlNet Auxiliary Preprocessors, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/Fannovel16/comfyui_controlnet_aux
pip install -r requirements.txt

Restart ComfyUI. Nothing to download - this is an image-manipulation utility, no model involved.

Where people get burned

First, don't mistake this for a preprocessor. It doesn't detect edges or estimate depth - it resizes a hint you already made. If you're feeding it a raw photo expecting a control map out, you've skipped the actual preprocessing step.

Second, the resize mode choice actually matters and picking the wrong one quietly hurts your result. Just Resize on mismatched aspect ratios warps your structure - a canny outline of a face gets subtly squished, and the generation follows the squish. If your ratios don't match, reach for Crop and Resize or Resize and Fill and decide whether you'd rather lose the edges or add borders.

Third, honestly: a lot of workflows never need this node, because they preprocess at the render's dimensions from the start. It earns its place when you're reusing a control map across different output sizes, or when your source and target aspect ratios genuinely differ. If everything's already the same shape, you can skip it.