Content Safety in AI Image Generation — What's Blocked and Why

Categories every major platform blocks

• CSAM — hard block, no exceptions. Reported to authorities.
• Non-consensual intimate imagery of real people.
• Deepfakes of political figures in most jurisdictions.
• Photorealistic weapons in a threatening context.
• Medical imagery claiming to diagnose or treat.
• Realistic depictions of minors in adult contexts.

Categories that are grey-area

• Public figures in ordinary contexts (varies by platform)
• Weapons in gaming / entertainment contexts (allowed on some)
• Nudity in artistic contexts (Midjourney: no; some SDXL fine-tunes: yes)
• Real brand logos (usually blocked)

Why the false positives

Safety classifiers are trained on patterns, not intent. A cute cartoon holding a stick can be flagged as a weapon. A child on a beach can be flagged as inappropriate.

If you're getting flagged on obviously safe prompts:

1. Rephrase — "toy sword" beats "sword"
2. Change the style — "cartoon illustration" instead of "photorealistic"
3. Remove the ambiguous word — the classifier is often keying on one token
4. If you have a business account, contact support with the flagged prompt

What Imagoat does

Imagoat runs standard safety filters at generation time. Legitimate business use — product photography, brand mascots, marketing — almost never triggers false positives because the prompt style is dramatically different from the abuse patterns the classifier was trained on.