-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy pathabout.html
More file actions
294 lines (278 loc) · 16.8 KB
/
Copy pathabout.html
File metadata and controls
294 lines (278 loc) · 16.8 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1" />
<title>Vision & About — Ghostmaxxing</title>
<link rel="icon" type="image/svg+xml" href="/images/ghostmaxxing-favicon.svg">
<meta name="description"
content="Ghostmaxxing is a browser-based public research platform for testing face-recognition camouflage, running workshops, and investigating real-world biometric surveillance." />
<meta property="og:title" content="Vision & About - Ghostmaxxing" />
<meta property="og:description" content="A public lab for testing face-recognition camouflage." />
<meta property="og:type" content="website" />
<meta property="og:image" content="/images/social-card.svg" />
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="Vision & About - Ghostmaxxing" />
<meta name="twitter:description" content="A public lab for testing face-recognition camouflage." />
<meta name="twitter:image" content="/images/social-card.svg" />
<link rel="stylesheet" href="styles/pages.css" />
<link rel="stylesheet" href="styles/content-pages.css" />
</head>
<body>
<main class="editorial-homepage content-page editorial-homepage--svg-footer-band">
<div class="editorial-homepage__shell">
<header class="editorial-homepage__header">
<a class="wordmark" href="index.html" aria-label="Ghostmaxxing homepage">
<span class="wordmark__main">
<img class="wordmark__mark" src="images/ghostmaxxing-mark.svg" alt="" aria-hidden="true" />
<span class="wordmark__title">Ghostmaxxing</span>
</span>
<span class="wordmark__subtitle">
A public lab for testing face-recognition camouflage
</span>
</a>
<nav class="editorial-homepage__nav" aria-label="Project navigation">
<a class="navlink" href="/references/" rel="noopener noreferrer">Research References</a>
<a class="navlink" href="about.html" aria-current="page">Vision & About</a>
<a class="cta-inline" href="report.html">How to <i>leak</i> to us<span aria-hidden="true">↗</span></a>
</nav>
</header>
<section class="content-page__header-area" aria-labelledby="page-title">
<span class="content-page__kicker">Vision & About</span>
<h1 id="page-title">Make the machine-readable face <i>contestable</i>.</h1>
<p class="content-page__lead">
Ghostmaxxing is a browser-based workshop assistant and public research platform for testing
face-recognition camouflage.
</p>
<p class="content-page__deck">
It grows out of privacy activism, adversarial makeup workshops, and the need to investigate which biometric
systems are actually used in public space.
</p>
</section>
<div class="content-page__layout">
<article class="content-page__main">
<section class="content-section" id="workshop-assistant">
<h2>Workshop assistant.</h2>
<p>
Designed to support group workshops, where participants play with makeup and learn how different faces
read to a recognition model: point a browser camera at your face, record a baseline, then change your
look — paint directly on the tracked face or follow a guided pattern — and watch how the
local recognition pipeline reacts in real time.
</p>
<p>
The goal is to lower the barrier to entry. You don't need a makeup specialist to take part — only
water-based colors, safe enough for kids. The system runs locally: it uses the webcam or a phone
camera, keeps everything on-device, and sends nothing anywhere else, except what you choose to share
with us.
</p>
<p>
The project is built to run in a phone browser, because makeup is usually applied in front of a mirror,
not at a desktop workstation.
</p>
<div class="callout">
<p>
Two kinds of success exist: the face stops being detected as a face at all, or the face is still
detected but its extracted points differ enough from the clean-face baseline that matching fails.
</p>
</div>
<h3>What can be sent to us?</h3>
<p>Three things mostly:</p>
<ol>
<li>
Before/after images of a successful attempt, sent via the
<!-- TODO: link/insert the share icon from lab.html once it's finalized -->
share icon in the lab.
</li>
<li>A one-second video of a successful attempt.</li>
<li>
A camouflage pattern you baked from a look that worked, so others can study and retest it. This is
the technical layer — mostly for people who want to look into how it works; see the
<!-- TODO: publish the Ghostyle authoring guide and confirm this link -->
<a href="/docs">Ghostyle documentation</a>.
</li>
</ol>
</section>
<section class="content-section" id="before-after">
<h2>Keep what works.</h2>
<p>
When a look defeats the local match, you can keep that exact state: paint it on the tracked face in the
lab and bake it into a reusable pattern others can study and retest. Automatically extracting a pattern
from a pair of before/after photos is a much harder research problem, and one we deliberately don't
depend on.
</p>
<div class="callout">
<p>
These approaches are experimental. We can't guarantee they will work — and a human reviewing
camera footage can often still recognize someone even when they're wearing makeup. A browser result
is a local, conditional finding, not a general protection claim.
</p>
</div>
</section>
<section class="content-section" id="origin">
<h2>Where it comes from.</h2>
<p>
Ghostmaxxing is the international evolution of an experiment run by NINA.watch. As part of the
Universal Digital Union, <a href="https://sindacato.nina.watch/it/iniziative/ghostati/">Ghòstati</a>
(<i>become a ghost</i>) is a workshop we keep repeating to explore adversarial makeup as a way to
resist facial recognition. The project began in May 2026 during the NINA festival, in
<a href="https://sindacato.nina.watch/it/blog/ghostati-workshop-maggio-2026/">Milan and Rome</a>. We
learned from <a href="https://www.michelletylicki.info/">Michelle Tylicki</a>, an artist who has run
this kind of workshop before us — we added the application development, and now this wider
vision.
</p>
<p class="callout">
Tylicki's <a href="https://www.michelletylicki.info/dazzle/">DAZZLE</a>, developed with Lauri Love, is
an art-and-tech installation and interactive tool for teaching Computer Vision Dazzle, also known as
anti-surveillance makeup. Ghostmaxxing inherits that workshop energy, but turns it into a browser-based
system people can run on their own devices.
</p>
</section>
<section class="content-section" id="vision">
<h2>Vision.</h2>
<ul>
<li>Make the workshop easier to run.</li>
<li>Make adversarial makeup a popular practice, not a niche one.</li>
<li>Collect evidence of what actually works.</li>
<li>Support new forms — clothing, 3D-printed objects, and beyond.</li>
<li>Support new research.</li>
<li>Use network effects to test everywhere, share results, and keep improving.</li>
</ul>
</section>
<section class="content-section" id="research-question">
<h2>The research question is minimal camouflage.</h2>
<p>
The core question is simple: how little makeup is needed to disrupt a face-recognition pipeline?
</p>
<p>
Less makeup matters because it lowers the barrier to practice. It makes the technique easier to learn,
easier to repeat, and easier to move from exceptional workshop performance toward everyday culture. The
hope is not to sell invisibility; it is to make adversarial makeup legible enough to become pop practice.
</p>
<h3>Success should be shareable.</h3>
<p>
Ghostmaxxing includes sharing-oriented functions because a successful look is also a teachable pattern.
People who want to share a one-second video, a before/after comparison, or a Ghostyle result can help
others understand what worked and what still needs to be tested.
</p>
</section>
<section class="content-section" id="leaks">
<h2>Why leaking matters.</h2>
<p>
The lab can test browser pipelines, but real-world facial recognition is a supply chain: camera
hardware, edge devices, model vendors, watchlists, matching systems, dashboards, alerts, metadata,
operators, procurement contracts, and retention rules.
</p>
<p>
<a href="report.html">Leak to us</a> exists to understand that hidden chain. Which technologies are
deployed? Who sells them? Who maintains them? What metadata is produced between capture and decision?
Where does it go? Who can access it? Tell us what you know, only if it is safe for you to do so —
and please don't include unnecessary personal data.
</p>
<p>
Some allies may be forced to work on or near these systems. If they can safely share information, they
may help us understand whether Ghostyles and adversarial makeup are actually affecting real
deployments, or whether the resistance needs to change.
</p>
<div class="callout">
<p>
In Europe, real-time remote biometric identification in publicly accessible spaces for law
enforcement is treated as a prohibited AI practice, subject to narrow exceptions and safeguards. That
legal frame still leaves a practical question: what is actually being deployed, and under whose
control?
</p>
</div>
<h3>A ban that keeps getting reopened.</h3>
<p>
The <a href="https://reclaimyourface.eu">Reclaim Your Face</a> campaign has spent years pushing for a
ban on biometric mass surveillance, and it helped get real-time remote biometric identification treated
as prohibited in EU law. But the fight didn't end with the text of the regulation: national security
agendas keep reopening the exceptions, arguing for carve-outs around major events, "special" cases, and
pilot deployments. <a
href="https://reclaimyourface.eu/biometric-surveillance-in-the-czech-republic-the-ministry-of-the-interior-is-trying-to-circumvent-the-artificial-intelligence-act/">Reporting
from the campaign</a>
has documented national authorities attempting to work around the AI Act's limits rather than comply
with them. A ban that keeps getting quietly worked around by fear-mongering, securitarian framing needs
the same kind of public pressure and evidence-gathering that got it written in the first place —
which is part of why the reporting node exists here too.
</p>
</section>
<section class="content-section" id="internal-tools">
<h2>Internal tools & diagnostics.</h2>
<p>
To support deeper analysis, model testing, and pattern sharing, Ghostmaxxing includes experimental diagnostic tools. These utilities run entirely locally in the browser to examine recorded behavior and transfer designs.
</p>
<h3>Video Loader</h3>
<p>
The <a href="loader.html">Video Loader</a> allows users to import local MP4 video files to run against our on-device 2D and 3D face-detection engines. This allows for precise frame-by-frame analysis, face extraction, database recording, and signature comparison without requiring a live camera feed. It is designed to evaluate recorded workshop outcomes or diagnostic video tracks under stable, repeatable conditions.
</p>
<!-- RELEASE: unreleased tools — uncomment on launch of Ghostyle Transfer / Latent Space Visualizer.
<h3>Ghostyle Transfer</h3>
<p>
The <a href="ghostyle-transfer.html">Ghostyle Transfer</a> tool extracts painted makeup patterns from a workshop before/after image pair and attaches them to a new target face. When the local face engine detects matching landmarks, the tool aligns the pattern using a 3D face mesh. Otherwise, it scales the pattern using manual bounding boxes, letting researchers visual-test camouflage layouts on different facial structures before physical application.
</p>
<h3>Latent Space Visualizer</h3>
<p>
The <a href="realtime.html">Latent Space Visualizer</a> is a real-time visual debugger for analyzing face descriptor drift. By recording a baseline face signature via webcam, it tracks facial embedding alterations and showcases biometric distance thresholds alongside a 128-dimensional descriptor equalizer. It is used to study exactly how specific facial movements, lighting shifts, and makeup strokes skew facial signatures.
</p>
-->
<!-- END RELEASE-gated tools -->
</section>
<section class="content-section" id="resources">
<h2>Read, test, report.</h2>
<p>
Ghostmaxxing is strongest when the lab, the archive, and the reporting channel work together. Test
techniques locally, read the lineage of anti-biometric appearance design, and help document the real
infrastructure when it appears in the world.
</p>
<div class="resource-grid">
<a class="resource-card" href="lab.html">
<h3>Open the lab</h3>
<p>Run local browser tests and watch a recognition pipeline react to your face in real time.</p>
</a>
<a class="resource-card" href="loader.html">
<h3>Video Loader</h3>
<p>Run pre-recorded MP4 video tests to extract and compare face signatures frame-by-frame.</p>
</a>
<!-- RELEASE: unreleased tools — uncomment on launch.
<a class="resource-card" href="ghostyle-transfer.html">
<h3>Ghostyle Transfer</h3>
<p>Extract makeup patterns from before/after images and project them onto new target faces.</p>
</a>
<a class="resource-card" href="realtime.html">
<h3>Latent Space Visualizer</h3>
<p>Debug biometric distance thresholds and visualize face descriptor drift in real time.</p>
</a>
-->
<a class="resource-card" href="/references/">
<h3>Research references</h3>
<p>Trace the artistic, technical, and activist lineage of face-obfuscation work.</p>
</a>
<a class="resource-card" href="/docs/">
<h3>Technical documentation</h3>
<p>Inspect modules, plugin logic, and implementation details.</p>
</a>
<a class="resource-card" href="report.html">
<h3>Leak to us</h3>
<p>Help investigate the real-world pipeline and supply chain of face recognition.</p>
</a>
</div>
</section>
</article>
<aside class="content-page__aside" aria-label="Page sections">
<h2>On this page</h2>
<nav>
<a href="#workshop-assistant">Workshop assistant</a>
<a href="#before-after">Gallery of success cases</a>
<a href="#origin">Origin</a>
<a href="#research-question">Research question</a>
<a href="#leaks">Why leaking matters</a>
<a href="#internal-tools">Internal tools</a>
<a href="#resources">Resources</a>
</nav>
</aside>
</div>
</div>
<div class="camera-band" aria-hidden="true"></div>
</main>
</body>
</html>