If you uploaded a photo you'd consider perfectly normal and got a "NSFW content detected" message, you're not alone. This article explains exactly where that check happens, what it's actually screening, and what to do if you think your photo was flagged unfairly.
Where the Check Happens
This is the part most people don't realize: the safety check does not run on the final result. It runs on your raw uploads, before anything else happens.
The actual request sequence, in order, is:
- You submit your source photo/GIF/video and your face photo.
- The server checks your daily free limit (or credit balance if you're signed in).
- Both files — the source and the face photo — are sent to an automated content-safety check.
- Only if both pass does the server upload your files to storage.
- Only after that does the AI generation model run.
In other words, the safety check is a gate that sits before upload and before generation — not a filter on the output. If either file trips it, the request stops immediately: nothing is generated, and nothing from that request is sent to the AI model. This is deliberate — it's designed to prevent unsafe content from ever reaching the generation step, not to catch it afterward.
What Gets Scanned
Both images are checked, every time — not just the one you'd expect. If you're doing a face swap, that means:
- The source photo, GIF, or video you're swapping into
- The face photo you're swapping in
A block on either file stops the whole request. This is why a face swap can fail even when the photo you assumed was the problem was actually fine — it might have been the other upload that tripped the check.
For video, a separate check runs against the whole clip rather than a single frame, since video carries more content across its duration than a still image.
What We Won't Publish (and Why)
We're not going to publish the specific detection threshold or describe how to construct an image that slips past the filter. Publishing that information would only help people deliberately evade a safety system that exists to prevent misuse — it wouldn't help anyone whose normal photo got flagged by mistake. What we can tell you is the category of content it targets (sexually explicit material) and, more usefully, what to do if you believe your photo doesn't belong in that category.
Why a Normal Photo Can Still Get Flagged
Automated content classifiers are not perfect, and false positives happen. Based on the pattern of blocked uploads we see, photos are more likely to trip the filter when they include:
- Swimwear, athletic wear, or other skin-revealing (but non-explicit) clothing
- Tight cropping around the torso or unusual framing that removes context
- Unusual lighting, filters, or color casts that distort skin tones
- Low resolution or heavy compression that obscures detail the classifier relies on
None of these guarantee a block — most photos in these categories pass fine — but they're the pattern behind most of the false positives we're aware of.
What to Do If You Think It's a False Positive
Since we don't expose the underlying detection logic or provide a manual review/appeal queue, the practical options are:
- Try a different photo of the same subject. A different angle, crop, or lighting condition often clears the check even when the subject is identical.
- Crop more conservatively. A tighter crop around just the face and shoulders, removing background or body context, is less likely to trigger a false positive than a full-body or loosely framed shot.
- Check the other file, not just the one you suspect. Remember — both the source and face photo are scanned. If your source photo is clearly fine, the face photo may be the one that was flagged.
- If neither upload should plausibly be flagged, contact support through the channels listed on the site rather than repeatedly retrying the same files — retrying an actually-blocked file will not change the result.
Photo and Video Are Screened Differently, But Never Skipped
Under the hood, our safety check uses a separate detection model for video than it does for still images (photo, GIF frame, and face photo uploads all go through the image-specific model). This is because video content needs to be evaluated across the whole clip, not a single frame, so it runs through a model built for that. The important part for you as a user is that the behavior is identical either way: a flagged file blocks the request, full stop, regardless of which mode you're using. Video face swap uploads get exactly the same treatment as photo uploads — there's no mode where the safety check is skipped or relaxed.
What This Filter Is Not
To be direct about scope: this is a pre-upload safety screen, not a moderation system that reviews or stores a record of your specific photo for human review, and it isn't a mechanism for detecting deepfakes or verifying consent between the people in your photos. It exists for one purpose — to stop explicit content from being processed — and every upload, from every user, goes through it before generation, with no exceptions and no way to bypass it from the client side.
Related Reading
Ready to try a swap? Use the face swap tool — photos and GIFs are free to try, no sign-up required.
