Inside Prompt.sex — How Our AI Identifies Nude Stars
Thousands of nude face searchs flow through Prompt.sex every day. A user uploads a photo; two seconds later, ranked results appear. But what happens in those two seconds? This article is a technical deep-dive into the pipeline that powers the entire platform.
Layer 1 — Detection (RetinaFace)
The uploaded image first hits a RetinaFace detector running inside an InsightFace wrapper. RetinaFace is a single-shot multi-scale face localiser that outputs bounding boxes and five-point landmarks (left eye, right eye, nose tip, left mouth corner, right mouth corner) for every face in the frame. If multiple faces are detected, the pipeline selects the one with the highest confidence score and the largest bounding-box area — the assumption being that the "main" face is the one the user intended to search.
Layer 2 — Alignment & Normalisation
Using the five landmarks, the detected face is affine-transformed into a canonical 112 x 112 pixel crop: eyes level, centred, with a consistent scale. This normalisation step is critical — it ensures that a face photographed at a 15-degree tilt and a face shot dead-on produce comparable inputs to the embedding network.
Layer 3 — Embedding (ArcFace)
The normalised crop passes through a ResNet-100 backbone trained with Additive Angular Margin Loss (ArcFace). The output is a 512-float unit vector — a point on the surface of a 512-dimensional hypersphere. Each dimension encodes a learned combination of facial attributes: no single dimension maps cleanly to "nose size" or "eye colour," but collectively they capture everything the network learned is useful for distinguishing faces during training on millions of labelled identities.
Layer 4 — Search (FAISS)
The query vector is fed to a FAISS (Facebook AI Similarity Search) index containing 800,000+ pre-computed embeddings — one per indexed thumbnail frame. FAISS performs an inner-product search, returning the top-K nearest neighbours in single-digit milliseconds. Because both query and index vectors are L2-normalised, inner product is equivalent to cosine similarity.
Layer 5 — Post-Processing
Raw FAISS scores are clipped, rescaled, and passed to the frontend. Results below a minimum quality threshold are dropped. The surviving matches are enriched with metadata (performer name, video title, source URL) and served as the final result grid.
Accuracy & Limitations
Several variables affect output quality:
- Input resolution: Faces below 112 x 112 pixels may not be detected at all
- Occlusion: Sunglasses, masks, or hands covering part of the face degrade embedding accuracy
- Expression: Extreme expressions distort landmark positions, shifting the resulting embedding
- Ageing: Our index spans performer photos from different years; a 10-year gap can reduce cosine similarity by 5–10 percentage points
The Performers
1. Jane Liddell
Jane Liddell, age 42, from . A recurring presence in scan results across multiple demographic segments, Jane Liddell holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
2. Hadas Yaron
Hadas Yaron, age 30, from Israel. A recurring presence in scan results across multiple demographic segments, Hadas Yaron holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
3. Han Seo-ah
Han Seo-ah, age 30, from . A recurring presence in scan results across multiple demographic segments, Han Seo-ah holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
4. Sabrina Impacciatore
Sabrina Impacciatore, age 41, from Italy. A recurring presence in scan results across multiple demographic segments, Sabrina Impacciatore holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
5. Angelica Hart
Angelica Hart, age 24, from . A recurring presence in scan results across multiple demographic segments, Angelica Hart holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
6. Rosamund Pike
Rosamund Pike, age 33, from United Kingdom. A recurring presence in scan results across multiple demographic segments, Rosamund Pike holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
7. Michele Goodger
Michele Goodger, age 46, from . A recurring presence in scan results across multiple demographic segments, Michele Goodger holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
8. Lesia Samaieva
Lesia Samaieva, age 33, from . A recurring presence in scan results across multiple demographic segments, Lesia Samaieva holds a distinctive position in our embedding index. The profile page offers full facial analysis and a porn lookalikes network that frequently surfaces unexpected cross-border matches.
Keep Exploring Porn Content Porn Content
76,000+ performers. 800,000+ indexed face embeddings. The surface has barely been scratched. Here are your next moves:
- Upload a nude face search photo and find your own nude celebrity double
- Browse female nude stars by category, ethnicity, or country
- Browse male nude stars for the other half of the index
- Live trending board — see who is surging right now
- A-Z nude star directory — find any performer by name
Female Nude Stars | Male Nude Stars | Trending Porn Stars | A-Z Nude Star Directory