NSFW Filter 2.0 API For Image Moderation

NSFW Filter 2.0 returns adult-content labels in categories.nsfw: nudity, sexual_activity, and sexual_suggestive. The separate categories.special_care group may contain sexual_minors for suspected sexualized content involving minors. Ordinary child photographs alone do not trigger that label. Treat every classification as a signal that needs context, rather than a verified fact about a person. With strict_special_care set to true, a special-care match independently makes flagged true. With false, special-care labels still appear but do not independently trigger that flag. This is an explicit behavior change to check when migrating from the older model, whose special-care setting strengthens normal concept matching. An NSFW Detection API can help prioritize a careful escalation process; it cannot establish age or make a final legal judgment. NSFW Filter 2.0 performs content detection without editing pixels or creating replacement images.

NSFW Filter 2.0 uses threshold, an optional number from 0.01 to 0.99 with a default of 0.5. Lower values make detection stricter. The old NSFW Filter instead uses threshold_offset from 0 to 0.1, in increments of 0.01, defaulting to 0.02. Do not copy the old parameter into a new request or assume these settings are interchangeable. NSFW Filter 2.0 also documents PNG/JPEG Data URLs and a 16 MiB limit; the legacy guide describes an accessible image URL and a 20,971,520-byte limit. Validate both dimensions and total pixels before submission. Test your NSFW Detection API settings against representative, lawfully handled examples and examine false positives alongside missed detections. These are documented interface differences, not evidence of a measured accuracy gain or a guaranteed performance improvement.

The NSFW Filter 2.0 API separates task execution from moderation outcome. A successful task has status succeeded, but its image can still be flagged. Read result.data.flagged together with categories.nsfw and categories.special_care; do not treat result: null as approval. The NSFW Detection API initially returns an accepted task, typically HTTP 202, while processing continues. A later HTTP 200 lookup only confirms that the query succeeded, so inspect status again. Use GET /v1/inferences/{id}, not the media-generation task route. Alternatively, send a top-level HTTPS callback_url. Verify webhook signatures, handle duplicate notifications, and acknowledge receipt with HTTP 2xx. NSFW Filter 2.0 gives your backend machine-readable signals for consistent routing and recovery.

Use NSFW Filter 2.0 at the point where your application decides to publish or distribute an image. A flagged result can hold a listing, restrict visibility, or create a human-review ticket according to your platform policy. A non-flagged result can allow the next workflow step, but does not prove universal safety. Keep failed checks distinct from completed classifications so network or preprocessing errors never become silent approvals. For marketplace catalogs, your application can schedule separate single-image requests; the NSFW Detection API does not take a batch image array. NSFW Filter 2.0 supports Automated Image Moderation while leaving sensitive judgments and escalation ownership with your team. All artwork here illustrates application scenarios and is not a real model test result.