permet de modifier de façon subtile et imperceptible une image, de sorte qu'elle ne puisse pas être rattachée à un visage sur une autre photo [...] : il retouche très légèrement les traits principaux du visage afin de tromper la reconnaissance faciale. Le procédé s'appuie sur une base de données contenant des visages de célébrités.

En utilisant des visages de stars ressemblant très peu aux photographies originales, les images s'en trouvent étrangement transformées.
«Les changements apportés à [mes] photos sont visibles à l'œil nu. Sur les images modifiées, j'ai l'air morte, ma fille de 3 ans a du duvet sur le visage et mon mari a l'air d'avoir un œil au beurre noir.»

Githup repo: https://github.com/fawkesrobotics/fawkes

Any gif that claims to be showing 60fps is simply untrue due to the fact that no web browser out there right now supports displaying gifs higher than 50fps. If you try to set your frame rate any higher than what the browser supports (or if you set the frame delay to zero) then most browsers will default to a playback of 10fps (over 5 times slower than you probably intended).

FROM: https://sebsauvage.net/links/?ZYOVJA

À savoir : les navigateurs ne supportent pas les GIF animés à plus de 50 images par seconde (50 fps).

Un outil comme celui-ci est utile pour fournir une version texte d'un fichier média (audio ou vidéo). L'objectif de ce programme est de fournir un outil simple et rapide pour permettre de récupérer un texte et l'utiliser comme version numérique.

Au commencement était le projet de communs numériques "Common Voice" initié par la fondation Mozilla [...] Puis vint le projet du logiciel libre Vosk, à l'initiative d'un collectif de chercheurs (Alpha Cephei) [...] Enfin, vint le projet de transcription de Tykayn, contributeur de l'association "April" [...] Pour finir, vint le projet Scribe, à l'initiative de l'équipe SI des Ceméa qui créa cette interface, simple et fonctionnelle, basique pour proposer aux utilisateur·rices de bénéficier de l'ensemble de ces fonctionnalités sous une forme plus conviviale.

Added

• new parameters new_x and new_y for cell() and multi_cell(), replacing ln=0, thanks to @gmischler
• new add_highlight() method to insert highlight annotations: documentation
• new offset_rendering() method: documentation
• new .text_mode property: documentation
• the page structure of the documentation has been revised, with a new page about adding text, thanks to @gmischler
• a warning is now raised if a context manager is used inside an unbreakable() section, which is not supported

Changed

• local_context() can now "scope" even more properties, like blend_mode: documentation

Fixed

• No font properties should be leaked anymore after using markdown or in any other situations (cf. #359), thanks to @gmischler
• If multi_cell(align="J") is given text with multiple paragraphs (text followed by an empty line) at once, it now renders the last line of each paragraph left-aligned,
  instead of just the very last line (cf. #364), thanks to @gmischler
• a regression: now again multi_cell() always renders a cell, even if txt is an empty string - cf. #349
• a bug with string width calculation when Markdown is enabled - cf. #351
• a few bugs when parsing some SVG files - cf. #356, #358 & #376
• a bug when using multi_cell(..., split_only=True) inside an unbreakable section - cf. #359

Deprecated

• The parameter ln to cell() and multi_cell() is now deprecated, use new_x and new_y instead.
• The parameter center to cell() is now deprecated, use align="C" instead.

Displaying deprecation warnings

DeprecationWarnings are not displayed by Python by default.

Hence, every time you use a newer version of fpdf2, we strongly encourage you to execute your scripts
with the -Wd option (cf. documentation)
in order to get warned about deprecated features used in your code.

This can also be enabled programmatically with warnings.simplefilter('default', DeprecationWarning).

A really nice JS-based image glitch effect, adding a hacker / horror mood...

If anyone wants to use it, I transformed that code into a Javascript function there:
https://github.com/jacksonbenete/email_terminal/blob/master/src/glitch-img.js

It has a dependency on the p5.js lib

It's MIT-licensed, as all public CodePens code

This year, researcher David Buchanan tried to implement parallel decoding using the iDOT information. During development, he made a simple programming mistake and ended up making a wonderful discovery. He could create a PNG file with platform-dependent rendering. It looked one way on Windows, Linux, Firefox, and Chrome, and a different way on a Mac with default Apple applications, like the Safari web browser. (He found that Apple had implemented the same bug!) Buchanan provided two sample pictures (Hello World and computers) to demonstrate this per-platform rendering. It didn't take long for other people to use his code and generate other examples. Many of these images were uploaded to FotoForensics:

All of these images were synthesized using two machine learning models, VQGAN and CLIP.

Create inspiring visual content in a collaboration with our AI enabled tools

1. DEEP STYLE
   The technique is a much more advanced version of the original Deep Dream approach. It is capable of using its own knowledge to interpret a painting style and transfer it to the uploaded image.

2. THIN STYLE
   This tool is a simplified version of the Deep Style one. It is not capable of creating advance transformations but it still shines with some exceptional results. It is also noticeably faster than its big brother.

3. DEEP DREAM
   Initially it was invented to help scientists and engineers to see what a deep neural network is seeing when it is looking in a given image. Later the algorithm has become a new form of psychedelic and abstract art.

Alt (Python & TensorFlow): https://www.tensorflow.org/hub/tutorials/tf2_arbitrary_image_stylization