Will your photo get flagged as
sensitive on X?

How this works

1. Same blueprint as the open-sourced model. X published the training code for its sensitive-media image model: an image embedding goes into a small dense network (BatchNorm → Dense 128/256, tanh or GELU → sigmoid), and thresholds are picked from precision/recall curves. Will It Flag? rebuilds that exact pipeline.
2. Open parts stand in for private ones. X never released its image-embedding model, trained weights, or data. We use OpenAI's open CLIP image encoder, project to 256 numbers like X's input, and train X's head on LAION's public CLIP embeddings. LAION's open, MIT-licensed detector runs as a cross-check.
3. Graphic violence is our own add-on. X's rules also cover gore, but that model isn't public, so we add a separately labeled zero-shot CLIP check.
4. Private by design. Everything runs in your browser (WebGPU, or WebAssembly as a fallback). Your photo is never uploaded, stored, or logged, and it is shown only inside a sealed preview frame that other scripts on this page, including ads, can't read. Nothing about your photo is sent anywhere.

Read the full guide: what X counts as sensitive media, how the check works, and its limits →

This is an estimate, not a guarantee. X's real system uses private models, more signals, and human review, and its decisions can differ. Not affiliated with X Corp.