Has This Ever Happened to You?

You spend hours generating images, experimenting with prompts, and putting together something worth sharing.

You upload it to Civitai, expecting people to see your work.

Instead, you've committed a vocabulary crime!

Perhaps you mentioned a Maine Coon cat. Perhaps you were describing Gothic Lolita fashion. Or maybe you used another perfectly ordinary word that happens to have an unfortunate second meaning.

And now you're expected to figure out which part of your 300-word prompt offended an automated moderation system.

Here are a few examples from other creators:

Apparently, making AI art now requires a working knowledge of every possible interpretation of every English word.

Which would be reasonable, of course, if the rules were consistently explained.

Introducing Civitai Crybaby

Since we can't reasonably expect automated moderation to understand every context, I figured we might as well have something that warns us about its potential misunderstandings.

Civitai Crybaby is a custom ComfyUI node that scans your prompts for words and expressions that could cause problems when uploading to Civitai.

And here's the interesting part: I used Civitai's own publicly available moderation resources to build it. Their open-source repository includes keyword lists, filtering utilities, and other moderation-related resources, which helped form the foundation of the detection system, alongside additional curated and user-reported problematic terms.

So yes, we're essentially using Civitai's own tools to warn us about Civitai. I thought that was rather poetic.

The node identifies potentially restricted terms, ambiguous expressions, and NSFW-related keywords, then tells you what it found and why it might matter.

Think of it as a spellchecker, except instead of fixing your grammar, it tells you which words might get you punished.

And since a simple text warning apparently wasn't enough, I also gave it colored alerts, highlighted keywords, notification history, and meme sound effects.

A very sensible use of development time, I'm sure.

Features

• Risk-Based Prompt Scanning

Detects potentially problematic expressions and classifies them into different warning categories.

• Dynamic Node Colors

The node changes color based on the highest detected risk level. Red for Critical, orange for High Risk, yellow for NSFW, blue for Ambiguous, and green when nothing suspicious is found.

• Highlighted Keywords

Detected words and expressions are highlighted directly in the node, so you can see exactly what triggered the warning.

• Colored Notifications and History

Displays notifications inside ComfyUI, with a history panel where you can review previous warnings.

• Professional and Sarcastic Modes

Choose between straightforward technical notifications and sarcastic commentary about the situation.

Both provide the same detection information, so the sarcasm is optional. Unfortunately, the moderation system isn't.

• Sound Effects

Four recognizable meme sounds, each assigned to a warning level

• Adjustable Warning Levels

Choose which categories of warnings you want to see, depending on how sensitive you want the scanner to be.

• Non-Blocking Operation

The node never stops generation, removes words, changes your prompt, or interferes with your workflow.

You can ignore every warning if you want.

It's a surprisingly liberating concept.

Installation

Extract the downloaded ZIP file into your ComfyUI/custom_nodes/ directory, then restart ComfyUI and refresh your browser. The Civitai Crybaby node should appear in your node menu.

I've also included a sample Krea 2 workflow with Crybaby already integrated, so you can load it into ComfyUI and see how everything works without setting it up yourself.

How Does It Work?

Civitai Crybaby uses a categorized keyword database informed by publicly available Civitai moderation resources, along with additional curated and user-reported problematic terms.

It examines your prompts, recognizes relevant words and expressions, and reports possible moderation or classification risks.

The scanner uses text normalization, word-boundary matching, phrase detection, and supported contextual exceptions to reduce false positives.

It also distinguishes between different kinds of findings.

For example, a keyword associated with NSFW classification isn't necessarily prohibited. Similarly, a word with multiple meanings shouldn't automatically be treated as a confirmed policy violation.

Sources

The project draws on publicly available resources, including:

• Civitai GitHub Repository

• Civitai Moderation Utilities

• Civitai Profanity Filter Documentation

• Public NSFW and whitelist dictionaries, alongside additional curated and user-reported terms.

The scanner operates locally and does not require sending your prompts to Civitai for evaluation.

Important: This is an unofficial tool. It doesn't have access to every internal moderation rule, and it cannot guarantee whether an upload will be accepted.

Civitai's moderation also involves mechanisms beyond keyword matching. Think of Crybaby as an early warning system, not a reproduction of the entire moderation process.

A Few Things Worth Knowing

Civitai Crybaby doesn't analyze the visual content of your images, determine policy compliance, or automatically correct your prompts.

Its job is simply to identify potential problems and give you enough information to decide what to do about them.

It also doesn't prevent you from using any words.

I thought that was a fairly important feature for a tool designed to help with censorship-related frustrations.

Final Thoughts

I understand the need for moderation, and I don't expect automated systems to get everything right.

But there is something particularly frustrating about spending hours creating something, only to discover that an unrelated word in your prompt has triggered an unexpected restriction.

This tool won't fix Civitai's moderation system. It won't eliminate false positives.

But it might help you identify a few potential problems before you discover them the hard way.

And if nothing else, I can now hear a perfectly timed BRUH whenever my vocabulary becomes a liability.

Progress, I suppose...