ComfyUI Node

EZ Detail

The face fixer — a detail pass in a single node

By Pauan·Created 8 months ago·Updated 5 months ago· 3
EZ Detail
  • image_settings
  • IMAGE
resize_type
scale_methodlanczos

EZ Detail is the pack's answer to the oldest itch in image generation: the face that's 90% right and 10% cursed. It adds a detail pass - upscale the region, regenerate at low denoise, downscale back - and it does it with two inputs and no graph spaghetti. The description in the source is blunt: "Generates more details, very useful for faces and eyes."

It operates on the image-settings bundle, so it slots in anywhere in the image chain: EZ txt2imgEZ DetailEZ Generate, or EZ img2imgEZ CropEZ DetailEZ Generate for a targeted face fix.

Inputs

  • image_settings (EZ_IMAGE_SETTINGS) - the image pipeline so far.
  • resize_type - a dynamic combo with three ways to scale the region before sampling:
    • scale total pixels - target megapixels (default 1.0 ≈ 1024×1024), aspect preserved. The "give me more pixels to work with" option.
    • scale by multiplier - scale factor (default 2.0). Classic 2× detail pass.
    • scale longer dimension - resize the long edge to a set value (default 1024).
  • scale_method (combo, default lanczos) - the interpolation used when resizing. The tooltip says it plainly: area is best for downscaling, lanczos for upscaling, nearest-exact for pixel art.

The output is IMAGE (EZ_IMAGE_SETTINGS) with the detail pass attached, wired into EZ Generate.

How it works

EZ Detail just stores the resize instructions on the settings. At generation time, EZ Generate crops your region (if any), resizes it up to the target size, runs the sampler, then downscales back to the original dimensions using area interpolation and composites the result into place. The trick that makes it "detail" rather than "upscale": because the sampler works at higher resolution and the result gets shrunk back down, fine features the model skips at low res get a chance to appear. Pair it with a low image_weight on EZ img2img and the regeneration is subtle - you keep the original intent, you just sharpen the details.

That last part is the real lever. A detail pass at image_weight 0.4–0.6 reads as "polish this," while 1.0 reads as "redo this completely." For faces you want the gentle end of that range - enough denoise to fix eyes, not enough to change the person.

Installing it

It ships with the Prompt Helpers pack. In ComfyUI Manager search "Prompt Helpers" (Pauan/comfyui-prompt-helpers), install, restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/Pauan/comfyui-prompt-helpers

Restart ComfyUI. No model downloads; the pack needs PyYAML and desktop-notifier and ComfyUI 0.8.0+.

If your detail pass comes out mushy or doubled, check the scale_method: defaulting to lanczos is right for upscaling, but if your workflow is downscaling a big source, area is the intended choice. And if the face isn't getting fixed at all, you probably skipped the crop - EZ Detail alone works on the whole frame, but the targeted face fix is EZ Crop + EZ Detail together, and that's the combination people actually reach for.

Categoryprompt_helpers/image

Inputs (3)

NameTypeDefaultDescription
image_settingsEZ_IMAGE_SETTINGS
resize_typeCOMBOSelects how to resize
scale_methodCOMBOlanczosInterpolation algorithm. 'area' is best for downscaling, 'lanczos' for upscaling, 'nearest-exact' for pixel art.

Outputs (1)

NameTypeDescription
IMAGEEZ_IMAGE_SETTINGS