The Porn Face Swap Encyclopedia A reference work · First edition · November 2026

Glossary of the practice

Technical and industrial terminology, alphabetically arranged. Terms marked ⚖ are treated fully in the Legal Lexicon.

The vocabulary of porn face swap borrows promiscuously from computer-vision research, forum slang and statute books. This glossary fixes each term to one definition, cross-referenced (see, see also, cf.) per house convention. Entries describe; they never instruct — no entry contains operational detail sufficient to reproduce the practice.

Reading the entries

Three vocabularies, one shelf

The terms below arrive from three directions and keep their accent: research vocabulary (embedding, latent space, inpainting) comes from the computer-vision literature; market vocabulary (churn, service economy, retention risk) describes the commercial layer as investigators document it; and statutory vocabulary (NCII, digital forgery — marked ⚖) comes from the legal texts where the practice is prosecuted. A reader should expect the three to collide: courts use "deepfake" where engineers use "composite," and the Disambiguation page exists precisely for the overlaps.

What entries never contain

Where each term leads

Cross-references use the house conventions — see points to the preferred term, see also to a related one, cf. invites comparison. Statutory terms carry the ⚖ mark and expand fully in the Legal Lexicon; mechanism terms link back to the headword entry where the taxonomy is enumerated.

A

Autoencoder swap n.
The original algorithmic face-swap architecture (c. 2017): a shared encoder compresses donor and target faces into a common latent space, then a donor-specific decoder reconstructs the donor's features onto the target's pose and expression. Training takes hours per donor and yields the highest-fidelity composites still produced. See latent space; cf. identity injection.
Artifact n.
(1) The composite image or clip produced by a swap. (2) A visible defect betraying synthesis — boundary seams, tooth smear, flicker. Forensic reading consists of cataloguing artifacts of the second kind in an artifact of the first.

B

Blending n.
The final compositing stage in which the rendered donor face is merged into the target frame — color transfer, mask-feathering and grain-matching included. Statistically the stage most likely to betray the composite; see boundary seam.
Boundary seam n.
The residual discontinuity at the edge of the swapped region — jaw line, hairline, ears — where donor rendering meets unmodified scene. The single most diagnostic forensic cue in still composites.

C

Churn n.
The domain-rotation cycle characteristic of the service economy: payment processors, hosts and registrars suspend operators, who re-register under fresh names and TLDs within weeks. The reason this encyclopedia defines archetypes rather than cataloguing brands.
Composite n.
An image or video constructed from two or more sources — in this field, donor face plus target body/scene. The genus to which every porn face swap artifact belongs; cf. wholly synthetic imagery.

D

Deepfake n.
Umbrella term (2017, from "deep learning" + "fake") for synthetic media depicting a real person's likeness. Porn face swap is its largest and oldest application. Legal usage varies — some statutes say "deepfake," others "digitally created intimate image" or "digital forgery." See Legal Lexicon, s.v. digital forgery.
Donor n.
The person whose facial identity is harvested for transfer — typically non-consenting, typically sourced from public photographs. In legal entries, the donor is the victim.

E

Embedding n.
A compact numeric vector capturing a face's identity features, extracted by a recognition model (ArcFace lineage). Embeddings are what identity-injection systems condition on — and why a single photo suffices as donor material.

F

Face reenactment n.
Driving a face's expression, gaze or mouth from another performance — related machinery, different operation: reenactment preserves the subject's identity while altering behavior. See Disambiguation.
Flicker n.
Frame-to-frame instability in swapped video — identity drift, brightness jumps, landmark jitter. The most reliable video artifact; its mitigation ("temporal smoothing") is the current arms-race front.

H

Hash-matching n. ⚖
Computing a perceptual fingerprint of an intimate image on the owner's device so platforms can block its upload without the image itself being transmitted — the mechanism behind StopNCII (adults) and NCMEC's Take It Down (minors). See Legal Lexicon.

I

Identity injection n.
Single-pass swapping via a pre-trained identity model: no per-donor training, one photo sufficient, seconds of compute. The technique behind the current volume of web-app and chat-bot output.
Inpainting n.
Generative filling of a masked region — the mechanism of "undressing" applications, which synthesize anatomy where clothing was masked. Related to but operationally distinct from face replacement; see Disambiguation.

L

Latent space n.
The compressed representational space in which autoencoder swaps encode facial structure; decoding a donor's latent representation through the target's pose is the swap itself.

M

Mask n.
The region selected for replacement — the face polygon in swaps, the clothing region in inpainting. Mask quality determines seam quality.
Morph n.
Continuous blend between two faces or images; occasionally confused with swapping but produces an averaged hybrid rather than a transferred identity.

N

NCII n. ⚖ (initialism)
Nonconsensual intimate imagery — the legal category under which porn face swap artifacts of real persons are prosecuted, alongside IBSA ("image-based sexual abuse") in international usage. See Legal Lexicon.

R

Retention risk n.
The exposure created when donor photographs, uploads and payment data persist in operator storage — unauditable "deletion" claims plus documented archive breaches make retention a defining hazard of hosted services.

S

Service economy n.
The commercial layer: credit-based web apps, chat bots, commissioned creators and hosting archives — characterized by churn, offshore processing and unauditable data handling. See churn.
Synthetic imagery n.
Wholly generated content depicting no real person — the category porn face swap is most often and most consequentially confused with. See Disambiguation.

T

Target n.
The source explicit material receiving the donor face; supplies body, scene, pose and performance. See composite.
Temporal consistency n.
Stability of a swapped identity across video frames; its absence is flicker, its engineering is the frontier problem of video pipelines.

U

Undressing application n.
A service that inpaints synthetic anatomy into a clothed photograph — the sister category to face swapping, sharing statutes, victims and enforcement routes. Cf. inpainting; see Disambiguation.

V

Victim n. ⚖
Legal term for the depicted person — almost always the donor, and the party whose consent is absent. Remedies attach to this status: hash-matching, the 48-hour federal rule, civil action. See Legal Lexicon.

Reading the taxonomy terms together

The mechanism family

Five entries describe how the artifact is made — autoencoder swap, identity injection, embedding, inpainting, latent space — and they hang together as one explanation: older pipelines trained a model per victim (autoencoder + latent space), newer ones reuse a universal identity model (embedding + identity injection), and the sister category synthesizes rather than transplants (inpainting). A reader who takes the five together has the technical shape of the whole field.

The forensic family

The second cluster — artifact, boundary seam, flicker, temporal consistency, blending — is the detector's vocabulary: each term names a place where the composite betrays itself. Detection research measures classifiers against exactly these cues; benchmark corpora like FaceForensics++ are built so that each artifact class can be scored separately.

The market family

The last cluster — service economy, churn, retention risk, donor — describes the industry around the tool: who operates it, why it keeps moving, and what an uploader actually surrenders. These entries carry the page's only warnings, because they are the vocabulary of the harm side, not the mechanism side.

Conventions: entries give preferred term, part of speech, definition and cross-references. Statutory terms are marked ⚖ and expanded in the Legal Lexicon; confusable pairs are resolved in Disambiguation; general orientation begins at the headword.