Why Are My AI-Generated Images Being Moderated?

If your influencer's images keep getting hidden by the moderation filter even though your prompt reads as totally clean, you're not imagining it. The filter doesn't just scan the words you typed; it also looks at the final rendered image, and sometimes at the prompt after our "magic prompt" feature expands it with extra descriptive detail.

Why a clean-sounding prompt still gets flagged

A few things can push a generation into the moderation queue without you ever typing anything explicit:

• Pose or camera-angle words ("leaning forward," "close-up," "arching back") combined with revealing outfits can read as suggestive to the filter, even if that's not your intent.
• Outfit terms like "sheer," "wet look," "skin-tight," or "cut-out" raise the risk score on their own.
• The magic prompt tool adds descriptive language to strengthen the image, and that added language can introduce words or phrasing the filter flags, even when your original prompt was fine. It's expanding your idea, not reading your intent.
• Lighting and background choices (certain bedroom or bathroom settings) nudge the risk score up regardless of what the character is wearing.

The model isn't reading your mind. It's scoring words and visual patterns, and sometimes combinations that seem innocent to a person land close enough to flagged patterns to get caught.

About the credits you lost

Generations that get hidden by moderation still consumed compute time to create, so the credit is spent whether or not you can see the result. That's expected behavior on our end right now, not a bug, though we know it's frustrating when it happens twice in a row on prompts that look harmless to you.

What to do about it

Start by removing adjectives describing body position, skin exposure, or fabric tightness, and test with a plainer version of the prompt first. If you're using the magic prompt feature, generate once with it and once without; if the plain version passes and the enhanced one doesn't, you've found your culprit and can edit the enhanced prompt manually before submitting.

If you believe a specific generation was wrongly flagged, contact support with the exact prompt text (copy it before you submit, not after) so the team can check what tripped the filter. That's the fastest way to get an answer on whether it's a wording issue or something with the model itself.