NSFW flags can come from Higgsfield's own safety system or from an external model provider's system, and both can produce false positives. Common safe-but-flagged triggers include swimwear, athletic content, medical imagery, certain reference images, and ambiguous prompt wording. Rephrase the prompt, adjust or remove the reference image, and retry. Credits for flagged generations are returned automatically for all models.
Why was my generation flagged as NSFW?
Higgsfield uses safety filters at two stages: during prompt processing and during image upload or reference input. External model providers (Nano Banana, Seedream, Kling, Sora 2, MiniMax, Wan, Veo) also have their own independent safety systems on top of Higgsfield's. Either system can produce a false positive.
Higgsfield doesn't have access to the internal moderation rules of external model providers, so the exact reason a specific generation was flagged is not always visible.
What commonly triggers false positives?
Swimwear or revealing clothing, athletic or fitness content, medical or anatomical imagery, reference images with borderline composition, and prompts containing ambiguous wording that safety filters may interpret as sensitive.
How do I fix a wrongly flagged generation?
Rephrase the prompt. Remove or replace words that could be interpreted as sensitive, and keep descriptions neutral and specific.
Adjust or remove the reference image. A borderline reference image can trigger the upload-stage filter even with a safe prompt. Try a different image with a clearer composition.
Combine image and text. If you used an image input without a text prompt, adding a descriptive prompt alongside it often reduces false positive flags.
Try a different model. Providers run their own safety systems, and a safe prompt that one system misreads may pass cleanly on another.
Will I get my credits back for a flagged generation?
For almost all models, credits for generations that fail due to NSFW flags are returned automatically. Verify in Manage Account → Usage: if you don't see the refund there, contact Support with the approximate time of the generation, the model you used, and your prompt. A flagged generation doesn't have a page in Assets, so there's no generation ID to attach: this information is enough for Support to locate it.
Higgsfield can't override safety classifiers manually: credits can be returned for confirmed false positives, but the classifier decision itself can't be reversed.