Why some generations get blocked
Every prompt and reference image is checked by an automated content safety filter before generation runs. When the filter refuses a request, no image comes back — but the generation still ran upstream, so the credit is used. This page explains what the filter is looking for and how to rephrase around it.
Why credits are consumed on a block
The safety check runs on the same infrastructure that generates the image, and we're charged for every completed inference — including refusals. Refunds are reserved for our-fault failures (network errors, invalid output, upstream server errors). A content-safety refusal isn't one of those, so the credit stays used. Rephrase and try again with the guidance below.
What commonly triggers a block
Real, identifiable people
Celebrities, public figures, and private individuals — by name or by photo reference. Likenesses of real people are almost always blocked, even in a pixel-art style.
Trademarked characters, logos, and brands
Named characters from games, films, or shows, brand logos, and other protected IP. Rephrase to generic archetypes ("plumber in red overalls" is not a workaround — describe an original character instead).
Graphic violence or gore
Detailed injuries, dismemberment, blood spatter, torture, or shock content. Stylized combat is usually fine; explicit gore is not.
Sexual or nude content
Nudity, sexual acts, or suggestive content — and anything sexualising minors, which is a hard block with no exceptions.
Hateful or extremist symbols
Slurs, hate symbols, and extremist iconography (historical or modern).
Reference images with any of the above
The safety filter inspects both your prompt and any reference or style image you attach. A clean prompt paired with a flagged reference will still be blocked.
How to unblock your prompt
Rephrase to generic archetypes
"Elven ranger with a longbow" instead of a named IP character. "Retro plumber hero" instead of a specific mascot. Describe the vibe, not the trademark.
Swap the reference image
Replace photos of real people or copyrighted screenshots with your own art, public-domain silhouettes, or a rough sketch. If a reference keeps getting blocked, generate without it and use the result as your new reference.
Tone down graphic descriptors
Drop words like "gory", "bleeding", "dismembered", or explicit anatomy. "Battle-worn knight" reads the same visually without tripping the filter.
Think it's a false positive?
The filter is deliberately conservative and occasionally flags harmless prompts — most often when a reference image has ambiguous content or when a fantasy descriptor overlaps with a real-world sensitive term. Try the rephrasing tips above first. If you're confident the prompt is safe and it keeps failing, send the exact prompt and (if applicable) the reference image to support so we can take a look.