🍒Bitch_Filter🎨母狗滤镜

Let's get the name out of the way: "Bitch_Filter" is a crass translation of a Chinese internet-culture term for the washed-out, warm, influencer-y photo look - and the author knows it. The code's own docstring says, in Chinese, "the filter name is 'Bitch Filter', for technical implementation only; actual deployment should consider renaming." So don't read anything rude into it; read it as "the internet-celebrity filter."

What it actually does is much more defensible: it emulates Kodak Gold 200 film - that warm, slightly faded, grainy look that reads as "shot on film" and, as the photorealism crowd keeps rediscovering, makes AI images look more like captured photographs. If your renders come out too clean and too glossy, this is a one-node way to knock them back toward something an actual camera might have produced.

How it works

The filter is a four-step pipeline: first it pushes the color balance warm (red up, blue down, with warmth controlling how far), then it fades the image toward white by fade to kill contrast, then - if add_grain is on - it adds monochrome noise so it doesn't touch hue. Finally the whole result is blended with the original by intensity, so 0 is untouched and 1 is the full Gold 200 look.

The grain is the detail that sells it: it's a single-channel (luminance-only) noise layer, which is exactly how film grain behaves - it grays the picture without tinting it. grain_distribution picks between uniform and gaussian; gaussian is the default and the more filmic of the two.

The inputs that matter

• intensity (0–1) - how much of the effect survives. Start at 0.6–0.8.
• warmth (0.8–1.5) - the yellow/red push. 1.5 is strong; 1.1–1.2 is usually enough.
• fade (0–0.3) - the black-lift. 0.2 default is a good starting point.
• add_grain + grain_strength (0–0.2) + grain_distribution - the texture layer. 0.07 is subtle; raise it for a heavy film look.

Output is a single IMAGE.

Installation

Standard KimNodes pack install - Manager → "ComfyUI_KimNodes", or git clone https://github.com/wjl0313/ComfyUI_KimNodes into custom_nodes, then restart. The filter needs cv2 (OpenCV) and torchvision; neither is in the pack's requirements.txt (which only pins pixeloe), so if the pack doesn't show up after install, pip install opencv-python torchvision and restart. torchvision ships with most ComfyUI portable installs, so usually you only need OpenCV.

Common issues

• The effect is too strong or too yellow - dial intensity down before touching warmth. The defaults lean heavy.
• Looks noisy rather than filmic - grain on a clean render reads as noise; lower grain_strength to ~0.03–0.05.
• Pack won't load - remember the pack's __init__.py imports every node at startup, so a missing OpenCV or torchvision takes down the whole pack, not just this filter. Check the terminal for the import error.

If you want a Kodak look as a finishing touch after your upscale pass, this is a reasonable one-stop shop - just don't type the node's name into a shared screen at work without a heads-up.