Como fazer (e identificar)

agot22

Scientists in Russia are claiming advances in deepfake creation that could make it easier to produce startlingly believable avatars.

Their techniques appear to make it possible to create a high-resolution deepfake video of a portrait using almost anyone as a movement model, or driver, and a single image of the image to be faked.

The researchers are from the Samsung AI Center in Moscow; the Skolkovo Institute of Science and Technology, also in Moscow; and Yandex, a Russian Internet-services company.

https://www.biometricupdate.com/202208/new-easier-way-to-make-deepfakes-emerges-from-russia

jul22

Researchers from Samsung Labs have developed a way to create high-resolution avatars, or deepfakes, from a single still frame photo or even a painting.

https://petapixel.com/2022/07/22/megaportraits-high-res-deepfakes-created-from-a-single-photo/

mai22

Researchers from Baidu and the South China University of Technology take this forward by developing the Head Swapper (HeSer), which can shift the entire head, not just the face. The key goal was to align the source head with the destination in a single blender that could cleanly handle both color and background mismatches.

https://www.marktechpost.com/2022/05/14/ai-researchers-from-china-propose-head-swapper-heser-for-few-shot-head-swapping-in-the-wild/

dez21

¿Recuerdas ZAO? Hace un par de años nos pilló por sorpresa no por ser la primera app en crear deepfakes fácilmente, sino por que el resultado era de gran calidad. Dos años más tarde, ZAO sigue sin salir de China, pero otras aplicaciones sí lo han hecho y ya están en lo más alto de las tiendas de aplicaciones, como es el caso de FacePlay.

FacePlay se puede descargar gratis en Google Play y la App Store, aunque te tocará pagar por las creaciones a base de ver anuncios y esquivar suscripciones. Con cuidado, podrás crear deepfakes increíbles de calidad que, como poco, te mantendrán entretenido durante un buen rato. https://www.xatakamovil.com/aplicaciones/faceplay-puedes-crear-increibles-deepfakes-movil-a-cambio-ver-algun-que-otro-anuncio

NOV21Disentanglement Is the Next Deepfake Revolution.

New research from Israel proposes a novel method of using synthetic data, such as CGI heads, to bring deepfaking into the 2020s, by truly separating facial identities (i.e. the essential facial characteristics of ‘Tom Cruise’, from all angles) from their context (i.e. looking up, looking sideways, scowling, scowling in the dark, brows furrowed, eyes closed, etc.). https://www.unite.ai/disentanglement-is-the-next-deepfake-revolution/

Out21
COMO SE FAZ (uma boa explicação)
I had the fortune of knowing a deepfake of me was coming; commissioned while researching my latest book, Trust No One: Inside the World of Deepfakes. Others won’t be so lucky. I told the deepfaker I wanted to see myself ‘commit’ a crime, but gave no further details: I wanted to be surprised so I could feel, as much as possible, how victims do. I provided an existing video of myself from the internet, and off they went. That last part is important because if a deepfaker wants to put your face into another video, they need to “train” their deepfake AI on thousands of images of yourself. This training is how the AI learns to copy your face exactly. But how could someone get thousands of images of you? It’s simple: if you’ve posted any video of yourself on the internet, the deepfaker can just grab the images from there. After all, every second of video is made up of about 30 photos. That means for every minute of video you’ve posted of yourself to social media, the deepfaker has 1,800 photos of your face.

 The AI then learns how to recreate your face and mask it over one in the video that the deepfaker wants you to appear in – a porno or a robbery, for example. And, exemplifying why deepfake tech is so frighteningly powerful, not only does your synthetic face get masked over the face of the original individual in the final video, but that individual’s expressions also get mapped onto your face. https://www.telegraph.co.uk/news/2021/10/30/deepfaked-committing-crime-should-worried/

TUTORIAL (sobre como se fazem e os riscos que apresentam)

https://www.youtube.com/watch?v=ccI0SCA04X4

============================== AULA 21=====

set21

Earlier this year, social media was briefly taken over by seemingly everyone using MyHeritage’s Deep Nostalgia feature to bring old photos to life. The company whose AI technology powers Deep Nostalgia, D-ID, is taking that technology one step further, turning still headshot photos into videos that move and say whatever a user wants. https://gizmodo.com/speaking-portraits-make-it-unsettlingly-easy-to-turn-st-1847766807

set21

Deepfake technology is shockingly sophisticated, allowing companies to create advertising clones, countries to imitate political rivals, and turn 50-year-old men into attractive young women. Now, livestreamers are using a new deepfake software to change their face in real time.  DeepFaceLive is an open-source AI software that can transform your face into someone else’s on video conferencing and streaming platforms, according to The Daily Dot. It was created by a developer known online as Iperov. Russian YouTuber Vladislav Pedro said that streamers have already started using the tool on sites like Twitch.  “DeepFaceLive has a lot of potential,” Pedro said to The Daily Dot. “This software can be used by both streamers and creators of any other media content.” https://futurism.com/the-byte/deepfake-livestreamers-real-time

jun21

COMO FAZER

Concrètement, l’IA Uberduck permet ainsi de choisir la voix d’une star de la musique puis de saisir le texte qu’elle doit prononcer. L’imitation est plutôt convaincante (mais pas toujours) et ce dispositif est très simple d’utilisation. https://www.presse-citron.net/deepfake-cette-ia-vous-permet-de-faire-un-duo-avec-votre-chanteur-prefere/ 

JUN221

(facebook combate e incentiva??)

O Facebook revelou um novo projeto que, aproveitando a tecnologia de IA da empresa, é capaz de imitar a escrita de uma pessoa bastando ter apenas uma palavra como base de estudo. Este sistema analisa a escrita do utilizador, e é capaz de converter qualquer outro texto que se pretenda para uma grafia similar.

Apelidado de TextStyleBrush, este novo sistema é capaz de reconhecer por IA o estilo de escrita de cada utilizador, bastante ter uma palavra como estudo. Feito isso, o sistema é capaz de replicar qualquer texto que se pretenda dentro dessa grafia, criando resultados verdadeiramente surpreendentes.

  Basicamente, podemos considerar esta tecnologia como um “deepfake” para escrita – o que pode ser visto como algo bom ou mau. O sistema não necessita de ficar limitado apenas a um estilo de escrita, é capaz também de reproduzir fontes.

Por exemplo, se possui um texto numa determinada fonte que pretenda aplicar noutro conteúdo, o TextStyleBrush é capaz de recriar a mesma – até se esta fonte for algo digital, e não propriamente uma escrita manual.

https://tugatech.com.pt/t39571-facebook-utiliza-ia-para-projeto-capaz-de-criar-deepfakes-de-qualquer-texto

jun21

A collaboration between Google AI researchers and the Indian Institute of Technology Kharagpur offers a new framework to synthesize talking heads from audio content. The project aims to produce optimized and reasonably-resourced ways to create ‘talking head’ video content from audio, for the purposes of synching lip movements to dubbed or machine-translated audio, and for use in avatars, in interactive applications, and in other real-time environments. unite.ai/googles-lipsync3d-offers-improved-deepfaked-mouth-movement-synchronization/

jun21

WOMBO CEO on app that allows people to easily make deepfakes on their iPhone

WOMBO is a lip-syncing app that has been downloaded 49-million times since launching earlier this year. Investors include Ashton Kutcher's venture capital firm. Ben-zion Benkhin, co-founder and CEO of WOMBO joins BNN Bloomberg to discuss the popular app, and why Canada is a good location for AI development. https://www.bnnbloomberg.ca/company-news/video/wombo-ceo-on-app-that-allows-people-to-easily-make-deepfakes-on-their-iphone~2213995 

abr21

4 Ways to Make a Deep Fake Photo, Video, or Meme

https://gadgetstouse.com/blog/2021/05/03/ways-to-make-deep-fake-photos-videos-memes/

abr21

Os 10 melhores websites e apps de Deepfake

https://www.showmetech.com.br/melhores-websites-apps-de-deepfake/

mar21

¿Quieres hacer un Deepfake para hacer reír a tus amigos? Si tu repuesta a esta pregunta ha sido “claro que sí”, queremos decirte que has llegado al lugar indicado. Aquí vamos a mostrarte cómo hacer un Deepfake en Telegram de forma simple y gratuita.

mar21
Agora, qualquer pessoa com um iPhone pode fazer um deepfake.

Agora, com apenas uma foto e zero experiência técnica, uma app para iPhone chamada Avatarify deixa-te controlar a cara da outra pessoa como um fantoche. Usando a câmara de selfie do teu telefone, o que fizeres com a tua cara vai acontecer na deles. A Avatarify não faz vídeos tão sofisticados e profissionais como os de Tom Cruise que percorreram o TikTok — mas foi descarregada mais de seis milhões de vezes desde Fevereiro. Outra app para iPhone e Android, chamada Wombo, transforma uma fotografia num lip-sync engraçado. Gerou 100 milhões de clipes apenas nas duas primeiras semanas.

E MyHeritage, um site de genealogia, deixa que qualquer um use tecnologia deepfake para trazer fotos antigas à vida. Basta fazer upload de uma foto de um familiar ou amigo falecido, e a app produz um pequeno vídeo muito convincente dele a olhar à volta e a sorrir. Até as pequenas rugas à volta dos olhos parecem reais. Chamam-lhe “Deep Nostalgia” e reanimou mais de 65 milhões de pessoas nas últimas quatro semanas.

https://www.publico.pt/2021/03/26/p3/noticia/qualquer-pessoa-iphone-deepfake-nao-prontos-vai-acontecer-1956134

mar21

O site MyHeritage, que oferece serviços voltados de genealogia online, lançou uma ferramenta que permite animar imagens usando deepfake. Chamada de "Deep Nostalgia", a plataforma tem como principal objetivo dar vida para fotos antigas, mas também pode ser utilizada em arquivos recentes. https://www.tecmundo.com.br/software/211808-site-permite-animar-fotos-antigas-usando-deepfake.htm 

MAR21

CONCEITO

Turek also highlighted a media manipulation technique called “deepfaking.” This is a process where the media manipulator trains two computer programs, called neural networks, using footage of a desired person, often a celebrity or politician, who they want to “fake” and footage of an actor whose facial expressions and movements will then be mapped onto the desired person’s face. In simple terms, deepfaking can make it look as if the desired person did or said something that they, in reality, did not. 

https://ndsmcobserver.com/2021/03/lecture-explores-deepfakes-media-manipulation/ 

DIFICIL

Ele usou um software de código aberto gratuito chamado DeepFaceLab como base para seus deepfakes e então empregou um poderoso hardware de processamento gráfico para rodar um sistema de inteligência artificial que foi treinado com mais de 13.000 imagens de Tom Cruise de todos os ângulos imagináveis do ator. Chris então criou detalhes adicionais com um conjunto de dados menor, de 5.000 a 6.000 imagens adicionais do ator. O vídeo demorou dois meses para ser concluído.  -  https://www.ofuxico.com.br/noticias-sobre-famosos/deepfake-de-tom-cruise-demorou-2-meses-para-ser-feito/2021/03/09-398102.html 

Fev21

Hour One’s promise to customers is that, after a relatively quick onboarding process, the company’s artificial intelligence can create a fully digital version of you, with the ability to say and do whatever you want it to. The company has partnered with YouTuber Taryn Southern to show off the tech’s capabilities. The biggest takeaway from this inside look is that Hour One’s clones are not deepfakes. A deepfake is created by manipulating an image to fabricate the likeness of a person, usually on top of existing video footage. Hour One’s clones, in comparison, require studio time to capture a person’s appearance and voice... and, therefore, consent. Southern says she stood in front of a green screen for about seven minutes, read a few scripts, and sang a song. This difference is noteworthy in that this capture process allows for a much fuller “cloning” process. Hour One can now feed just about any script into its program and create a video where it appears that Southern is actually reading it. There’s also an extra layer of consent involved — deepfakes are often made without the subject’s approval, but that’s not possible with Hour One’s technology. https://www.inputmag.com/tech/were-begging-you-to-not-turn-yourself-into-ai-powered-clone

Jan29

LIST:https://wwwhatsnew.com/2021/01/02/algunas-opciones-para-crear-deepfakes-en-video/

Jan20

DETEÇÂO

“In the short term, detection will be reasonably effective,” says Subbarao Kambhampati, a professor of computer science at Arizona State University. “In the longer run, I think it will be impossible to distinguish between the real pictures and the fake pictures.”[2] LINK

MASTER

Deepfakes affecting reputation

a study comparing effects of different levels of (fake) media on reputation

Nov20

“There are now businesses that sell fake people. On the website Generated.Photos, you can buy a “unique, worry-free” fake person for $2.99, or 1,000 people for $1,000. If you just need a couple of fake people — for characters in a video game, or to make your company website appear more diverse — you can get their photos for free on ThisPersonDoesNotExist.com. Adjust their likeness as needed; make them old or young or the ethnicity of your choosing. If you want your fake person animated, a company called Rosebud.AI can do that and can even make them talk.” LINK + The New York Times this week did something dangerous for its reputation as the nation’s paper of record. Its staff played with a deepfake algorithm, and posted online hundreds of photorealistic images of non-existent people. For those who fear democracy being subverted by the media, the article will only confirm their conspiracy theories. The original biometric — a face — can be created in as much uniqueness and diversity as nature can, and with much less effort. LINK

nov20

mais uma app

There are literally thousands of facial recognition tools out there that populate their data pools by collecting people’s photos on social media, training their machine learning algorithms, and making connections between faces and identities. If that sounds scary, you may use an artistic avatar or something that has nothing to do with your face. Alternatively, you can use ‘Anonymizer.’ This is an AI-powered deepfake generator that can create images that are close to resembling your facial characteristics, but not quite. The idea is that you can still use a lifelike image that is not stolen from another person, doesn’t violate any copyrights, and will make your profile look realistic to other users. All you need to do is to upload a clear photo of your face, and two separate generative adversarial networks will spew out 20 deepfakes. https://www.technadu.com/anonymizer-tool-helps-generate-deepfakes-your-face/227855/

ou20 COMO IDENTIFICAR (MUITO BOM)

https://edition.cnn.com/interactive/2020/10/us/manipulated-media-tech-fake-news-trnd/

ou20

Ele cita, por exemplo, os sistemas de deep learning capazes de criar rostos com aparência real, mas que não pertencem a pessoas de verdade. Tais plataformas usam uma tecnologia chamada de Rede Geracional Adversarial, que funciona, basicamente, em duas etapas: na primeira, o algoritmo gera uma imagem humana que julga verossímil, enquanto uma segunda IA, com base em fotos reais, avalia se a foto realmente passaria como a de um ser humano. Caso contrário, todo o processo é recomeçado, com o resultante sendo as imagens que passarem com sucesso pelos dois crivos virtuais. O mesmo processo também é usado na criação dos deep fakes, de forma que a transposição do rosto de uma pessoa para o corpo de outra aconteça, também, da forma mais natural possível. E assim como as fotografias de pessoas que não existem, no caso de um resultado bem feito, podem ser usadas de forma altamente danosa. “Estamos vendo uma verdadeira armamentização das [imagens falsas], que podem ser usadas para atingir a imagem dos indivíduos reais, manipular a opinião pública ou praticar crimes”, afirma Farid. https://canaltech.com.br/inteligencia-artificial/sociedade-pode-estar-perdendo-a-guerra-contra-os-deep-fakes-alerta-professor-172775/

ou20

 it seems that Nvidia's getting in on the holiday cheer with its newly announced Nvidia Maxine video streaming platform, which (among other features) essentially lets you video chat while wearing a deepfake style mask of...yourself 

https://www.tomshardware.com/news/nvidia-maxine-is-helpful-but-unsettling 

ou20

lista de opções

https://www.tecmundo.com.br/software/204602-deepfake-app-celular-coloque-rosto-gifs-videos.htm

set20

Synthesia es una startup que hace unos días presentó una herramienta que permitirá crear vídeos presentados por una "persona" que puede hablar 38 idiomas. Como se observa en el clip de presentación, una vez que se ha escrito el texto se puede elegir el idioma y entre varios modelos de "actores" https://www.iproup.com/innovacion/17031-synthesia-asi-podes-crear-videos-deepfakes-con-ia

set20

En esta ocasión, revisaremos una herramienta que permite editar clips con esta técnica, obteniendo buenos resultados. Además se puede probar directamente desde la web, aunque sólo con clips cortos. Se trata de Wav2Lip, una herramienta disponible en GitHub como parte de una investigación publicada en un paper titulado “A Lip Sync Expert Is All You Need for Speech to Lip Generation In the Wild”. https://wwwhatsnew.com/2020/09/16/wav2lip-una-herramienta-que-simplifica-la-creacion-de-deepfakes/

set20

I learned to make a lip-syncing deepfake in just a few hours (and you can, too) https://www.theverge.com/21428653/lip-sync-ai-deepfake-wav2lip-code-how-to 

set20

Cómo crear tus propios deepfakes (VARIAS APLICAÇÕES)

https://hipertextual.com/2020/09/apps-gratuitas-crear-deepfakes 

set20

Os vídeos deepfake estão a aumentar, e a contribuir para isso está o facto de ser cada vez mais simples criar vídeos manipulados. Este tutorial de como criar um deepfake para gerar vídeos Baka Mitai - um meme que se popularizou com os deepfakes, que já foi aplicado ao Obama e até ao Thanos - é um exemplo disso, resumindo-se a 8 passos: https://abertoatedemadrugada.com/2020/08/como-criar-um-deepfake-baka-mitai.html

ago20

Within a day, she had created a step-by-step YouTube tutorial to walk others through the process. “Making one of these deepfakes and overlaying audio is not as complicated as you may think,” she says in the video, published on August 4. It has since been viewed over 360,000 times. The particular deepfake algorithm that people were using comes from a 2019 research paper presented at NeurIPS, the largest annual AI research conference. Unlike other, more complex algorithms, it allows a user to take any video of a person’s face and use it to animate a photo of someone else’s face with only a few lines of code. Windheim found the open-source algorithm in a YouTube tutorial and ported it into a Google Colab notebook, a free service for running code in the cloud. After a few tries, aided by the skills she’d picked up in the occasional coding class in college, she got the script to spit out a deepfake video. She then synched the song to the video with Kapwing’s tools, creating a new version of the meme. https://www.technologyreview.com/2020/08/28/1007746/ai-deepfakes-memes/

ago20

As deepfakes podem ter bastantes objetivos no final, mas nem sempre têm de ser negativos. Um dos claros exemplo disso pode encontrar-se na recentemente lançada aplicação “Reface”. Esta nova aplicação, disponível para Android e iOS, utiliza o poder da Inteligência Artificial para criar vídeos deepfake, mas num formato mais social e divertido. A aplicação permite capturar uma selfie do rosto dos utilizadores, e colocar a mesma em diversos vídeos de atores famosos, cenas do cinema e centenas de outros efeitos. A aplicação possui o mesmo conceito que a reconhecida “FaceApp”, mas invés de aplicar o efeito apenas em fotos estáticas, os utilizadores podem colocar os seus rostos em praticamente qualquer cena animada em vídeo. https://tugatech.com.pt/t34726-reface-utiliza-ia-para-criar-deepfakes-com-rosto-dos-utilizadores

A facilidade e velocidade do aplicativo RefaceAI são seu cartão de visita. E isso é em parte o que permitiu que a empresa ganhasse terreno tão rapidamente: criando conteúdo de lançamento de produto para empresas como a marca de beleza Vivienne Sabo, fazendo parceria com John Legend e Miley Cyrus para vídeos promocionais e alcançando consumidores comuns que ganharam 20 milhões de downloads a RefaceAI e uma taxa viral de 65%.

https://gq.globo.com/Noticias/Tecnologia/noticia/2020/09/deepfake-voce-tambem-pode-ser-o-proximo-rosto-da-gucci.html

Ukrainian startup RefaceAI has developed mobile application Doublicat that can place its users’ faces on animated images of celebrities (known as GIFs) and make them move realistically. All one needs to do is to take a selfie and choose a GIF. https://www.kyivpost.com/technology/ukrainian-startup-creates-deepfakes-for-fun.html?cn-reloaded=1

¿Quieres sentirte famoso? Así puedes poner tu rostro en escenas de películas. https://www.unocero.com/como-se-hace/refaceapp-como-funciona/

ago20

In a series of videos posted to YouTube this week and shared by Motherboard, Attenborough’s sonorous voice is used to questionable effect by being assigned to to an A.I. that reads out Reddit threads. The result is Reddit with a whole lot more gravitas than you’d normally expect from it, all thanks to a little deep learning magic. https://www.digitaltrends.com/news/audio-deepfake-david-attenborough/

ago20

Reportagem apurou que aplicativo pretende lançar rostos de personalidades brasileiras; o Impressions App já permite simular imagem de Trump e Obama, além de celebridades dos EUA https://apublica.org/2020/08/yes-nos-temos-deepfake-brasileiros-sao-o-2o-maior-publico-de-aplicativo-que-troca-rostos-de-politicos-e-celebridades/ + https://outraspalavras.net/outrasmidias/yes-nos-temos-deepfake/

ago20

Fácil acesso a softwares para criação de deepfakes preocupa pesquisadores. Principal uso dessa tecnologia atualmente é para criação de vídeos de humor, mas ela já foi usada para fraudes e fake news https://olhardigital.com.br/noticia/facil-acesso-a-softwares-para-criacao-de-deepfakes-preocupa-pesquisadores/104767

agos20

Deepfakes Are Getting Better, Easier to Make, and Cheaper. 

GitHub is becoming a destination site for make-your-own-deepfake software. They mention a generative adversarial [neural] network or GAN model dubbed StyleGAN2, the underling code of which is available on GitHub. Tully and Lee showed that with a dataset of compiled Tom Hanks pics, structured to be of the same size and general direction, they could easily use StyleGAN2 tools to create new, convincing images of the actor, which they revealed in their Wednesday demonstration. 

https://www.defenseone.com/technology/2020/08/deepfakes-are-getting-better-easier-make-and-cheaper/167536/

jul20

It’s simply a piece of AI software called Vocodes. The tool, which I can best describe as a deepfake generator, can mimic the voices of a slew of politicians and celebrities including Donald Trump, Barack Obama, Bryan Cranston, Danny Devito, and a dozen more. All you need to do is write anything, choose a voice you’d like to mimic, and Vocodes takes care of the rest. https://thenextweb.com/neural/2020/07/28/celebrity-voices-deepfake-ai-app/

jul20

There’s an incredible new mobile app that lets you put your face onto famous movie scenes, Netflix shows, HBO shows, and even video game characters. The free version is a bit limited, but it lets you plonk your likeness on characters like Geralt, Joel from The Last of Us, Agent 47, Johnny Silverhand, and more. You can even be a Little Sister (reader, I made myself a Little Sister). The app is called Doublicat and the free version makes you watch some adverts between putting your face onto people. https://www.vg247.com/2020/07/23/mobile-app-deepfakes-face-onto-video-game-characters/

jul20

The deepfakes, Those manipulated videos that change the face of one person for another and superimpose the voice, reach video calls. This function is possible thanks to Avatarify, an open source software that uses artificial intelligence to change your face for another person’s (like a famous person) in real time, reproducing the gestures and movements you make. As reported by VICE, this software has been developed by programmer Ali Aliev, who posted a video on his YouTube channel a few weeks ago, showing how he made his friends believe he was Elon Musk in a video call via Zoom. https://www.thebulletintime.com/news/this-software-allows-you-to-use-deepfakes-in-video-calls/

jul20

Um sistema de inteligência artificial desenvolvido pela Microsoft em parceria com pesquisadores da Universidade de Zhejiang, na China, conseguiu gerar vozes cantantes em diferentes idiomas. Apelidado de DeepSinger, ele abre caminho para futuros deepfakes musicais. https://olhardigital.com.br/noticia/-inteligencia-artificial-da-microsoft-imita-vozes-cantantes-e-pode-criar-deepfake-musical/103455

jul20

Recently, a team of researchers from UC Berkeley and Adobe Research proposed a new machine learning model known as Swapping Autoencoder, which has the capability to perform image manipulation. The key idea of this research is to encode a picture into 2 independent components and then enforce that any swapped combination maps to a realistic image. https://analyticsindiamag.com/latest-model-that-might-replace-gans-to-create-deepfakes/

ul20

Os deepfakes são aqueles vídeos criados por inteligência artificial que imitam perfeitamente uma pessoa, e fazem acreditar que ela esteja realmente sendo retratada naquelas imagens. Por mais surpreendente que alguns deepfakes sejam, eles ainda têm uma certa limitação de resolução, mas um novo avanço dos estúdios da Disney pode acelerar a chegada da técnica às produções cinematográficas. Atualmente, um deepfake, por mais bem feito que seja, não consegue criar imagens com resolução superior a 256 x 256 pixels quando a ferramenta DeepFakeLab é usada. A tecnologia da Disney, em comparação, é capaz de criar imagens com 1024 x 1024 pixels.

https://olhardigital.com.br/noticia/disney-desenvolve-tecnica-de-deepfake-que-cria-imagens-em-alta-resolucao/102808 

Disney is committed to making deepfakes more realistic than ever with higher resolution, and its face-swap tech is closer to being used in big-budget movies and TV. In a recently published paper and video, Disney Research Studios explains how it has used progressive algorithm training, stabilisation technology and lighting effects to achieve convincing face-swapping results. Disney researchers also debuted the first photo-realistic deepfake at a megapixel resolution. Deepfake technology usually focuses on smooth facial transfers, rather than increasing the pixel count, which means that traditional deepfakes have more flaws on a larger screen. https://blooloop.com/news/disney-deepfake-face-swap-technology/ 

jun20

In the video above, Dr. Károly Zsolnai-Fehér‘s explains that anyone can now create deepfakes. In order to use this technique, he explains that all you have to do is film yourself and add the face of your choice. In a deepfake, several movements are transcribed, such as those of the head, mouth, or eyes. But it is also possible to move the whole body. https://todaynewstalk.com/2020/06/22/anyone-can-now-create-deepfakes-with-a-quality-that-is-beyond-comprehension/

jun20 

DeepFaceDrawing is a new AI system designed to generate realistic images of human faces based on simple or incomplete sketches. Created by a team of researchers at the Chinese Academy of Sciences and City University of Hong Kong, their 

research paper

 proposes that it can be used for anything from character design to criminal investigations. https://www.scmp.com/abacus/tech/article/3089833/chinese-ai-system-uses-deepfake-technology-turn-drawings-realistic

maio20
Tips for identifying deepfakes. https://www.professionalsecurity.co.uk/news/interviews/tips-for-identifying-deepfakes/

Abr20
Los 'deepfakes', uno de los peligros de la red más difíciles de detectar https://www.rtve.es/noticias/20200418/deepfakes-bulos-peligro-internet-coronavirus/2012176.shtml
1br20
Diferente de outros geradores de deepfakes, o Avatarify não precisa ser treinado com a imagem alvo e pode funcionar em tempo real, se o PC tiver uma GPU poderosa o suficiente https://olhardigital.com.br/noticia/software-aberto-cria-deepfakes-em-tempo-real-durante-reunioes-no-zoom/99619

abril20

But what if anyone could create a decent-looking deepfake right from their mobile device? A new app for the iPhone known as Impressions is aiming to do just that.

The app, which launched in select countries in late February before its global release last week, lets users transform themselves into an array of celebrities with just a few clicks. https://www.dailydot.com/debug/impressions-deepfake-app/ + https://metro.co.uk/2020/04/07/celebrity-deepfake-videos-12520788/

5 aplicaciones para hacer deepfakes en Android https://androidphoria.com/aplicaciones/mejores-aplicaciones-deepfakes-android

Existe uma biblioteca de código aberto, que utiliza a GPU para fazer o treinamento (da AI). Eu forneço milhares de imagens da pessoa que eu quero copiar, ela analisa essas imagens e depois é capaz de reproduzir".Depois, vem a parte da edição. "Só o deepfake não te entrega uma coisa tão real, no pós-produção eu já consigo melhorar". Ele leva cerca de sete dias para fazer um vídeo. "Varia muito. Para treinar o rosto estou levando uns quatro dias e depois mais uns três dias para terminar o vídeo todo". https://www.correio24horas.com.br/noticia/nid/verdadeiro-ou-falso-videos-deepfake-estao-cada-vez-mais-reais/

Técnica GAN é a mais popular

Existem várias técnicas para produzir um vídeo usando deepfake. De acordo com a pesquisa, a mais popular é a GAN, sigla para Generative Adversarial Network, famosa pela variedade de possíveis aplicações e resultado realístico. Uma das pistas da alta demanda pelo método é quantidade crescente de artigos científicos publicados sobre o tema: desde sua invenção em 2014, trabalhos que mencionam a técnica aumentam de quantidade exponencialmente e já registram 1207 publicações apenas em 2019.

https://www.techtudo.com.br/listas/2019/10/96percent-dos-videos-de-deepfake-tem-conteudo-pornografico-veja-sete-fatos.ghtml

19/11/2019 El vídeo es obra de Face To Fake, una misteriosa cuenta de Youtube y Twitter que hasta el momento no ha concedido entrevistas con ningún medio que anunciaba el pasado 12 de Julio su irrupción en la escena mediática con un avatar cuyos colores recuerdan a la bandera republicana: "¡Saludos a todos! Este es el inicio de una larga serie de vídeos basados en el arte del Deepfake, son muchas las ideas y espero que solo sea una cuestión de tiempo poder llevarlas a cabo. Disfrutad, ¡y nos vemos las CARAS!". En la descripción del vídeo del Equipo E aparece un agradecimiento que reza: "Agradecimientos a Iperov, quien puso esta magia al alcance de todos". Con un enlace a una cuenta del repositorio de software de código abierto Github, donde se encuentra el DeepFaceLab, el software que ha hecho posible el vídeo del Equipo E.

07/11/19 O algoritmo criado pela Nvidia se diferencia das atuais IAs usadas em vídeos deepfakes especialmente em dois pontos: o primeiro deles é que, enquanto as IAs atuais necessitam de uma grande base de dados (vídeos ou fotos) da pessoa ou objeto que se quer introduzir digitalmente em um vídeo, a nova IA da Nvidia consegue fazer isso com apenas uma única foto da pessoa. A segunda diferença está na capacidade de alterações possíveis: enquanto as IAs atuais apenas conseguem fazer apenas a troca de objetos já presentes no vídeo (por exemplo, trocar um rosto pelo outro), a da Nvidia consegue, a partir de uma foto, criar até mesmo movimentos realistas baseado na imagem base. LINK

É deepfake? https://stockdailydish.com/deepfakes-in-action-seattle-tv-station-accused-of-doctoring-trump-speech-video/

como FAZER

6/11/19 “You do not need to be a software developer to use this kind of technology,” West said. Creating your own deepfake video is simple, in theory. You would start with two real videos featuring two real people, and then take those clips and feed them into a series of computer algorithms. The technology breaks each source down frame by frame, and is widely available online. LINK

Clonar a voz e por-nos a dizer o que nunca dissemos “Software analisa uma amostra da voz original e a converte em um modelo matemático, que pode ser aplicado a um sistema capaz de ler qualquer texto”

https://olhardigital.com.br/noticia/inteligencia-artificial-e-capaz-de-clonar-sua-voz-em-5-segundos/92990

ByteDance has secretly built a deepfake maker, with the code for the feature hidden in the latest version of its massively popular video app TikTok. The secret feature was discovered by Watchful.ai, according to TechCrunch. The Israeli in-app market research startup found the code in the Android versions of TikTok and its Chinese sister app Douyin. Watchful was able to activate the code in Douyin to test the feature, which is not yet available to the public. LINK

Snap has acquired San Francisco based deepfakes startup named AI Factory.. Snap’s motive behind this acquisition is to use the deepfakes technology of AI Factory to power its recently launched Cameos feature. Cameos had launched in mid-December. With this feature, users can insert their selfie deepfakes-style into any desired scene, and it will create a short looping video, which can be shared via chat. Snapchat users can imitate themselves into a singing bird wearing a chef’s hat, a flying hotdog, a dancing roast chicken, a talking cat, and a lot more. LINK + https://www.snap.com/pt-BR/news/

Jan20 “mais um app surge para usuários de Android e iOS: o Doublicat.. Usando como base o RefaceAI, o aplicativo se diferencia por gerar GIFs sem a dependência de utilizar um modelo de rosto 3D como é feito pelo Zao, usando apenas conteúdo coletado e pré-processado, além de poder ser instável quando se consideram expressões faciais e curvas. LINK

JAN20 Researchers from Nanyang Technological University in Singapore, the National Laboratory of Pattern Recognition in China, and artificial intelligence software company SenseTime developed the method for creating deepfakes from audio sources. Basically, the AI takes an audio clip of someone speaking, and a video of another person (or the same person), and generates realistic footage of the person saying the words from the source audio. The person in the video becomes a puppet for the original voice. LINK

JAN20 este año ya han salido a la luz nuevas aplicaciones para crear “deepfakes”, que se han vuelto virales en apenas unos días. A continuación te detallamos las dos aplicaciones con las que podrás crear curiosos “deepfakes” en apenas segundos. La primera de ellas es Morphin, un creador de GIFs de nueva generación que permite a los usuarios infiltrarse en sus GIFS favoritos, convirtiéndote en el centro de atención. La nueva versión cuenta con etiquetas de búsqueda, GIFs en alta resolución y un aspecto más elegante. A diferencia de otro tipo de apps de este estilo, Morphin utiliza la tecnología CGI para recrear completamente la imagen el usuario. (...) En lo que llevamos de año también ha salido a la luz Doublicat, una aplicación que permite a los usuarios personalizar GIFS cambiando sus propios rostros por los de los protagonistas de dichos GIFs. La app pide nada más comenzar que te tomes un selfie, luego escanea la cara y la cambia por el protagonista de la escena del GIF seleccionado LINK

FEV20 A aplicação que hoje sugerimos é um gerador de vídeos onde pode usar o seu rosto para entrar nas cenas que quiser, mas não se trata de deepfakes mas de um simulador que inclui a criação de um modelo 3D do seu rosto. Chama-se mesmo Face em 3D - colocar seu rosto em um vídeo e faz o que promete, permitindo a criação de gifs e vídeos, que pode partilhar com amigos em animados memes. https://tek.sapo.pt/mobile/android/artigos/esqueca-os-deepfakes-e-coloque-o-seu-rosto-em-animacoes-para-partilhar-com-os-amigos

FEV20 Deepfakes use different underlying AI technologies—notably autoencoders and generative adversarial networks (GANs). An autoencoder is an artificial neural network trained to reconstruct input from a simpler representation. A GAN is made up of two competing artificial neural networks, one trying to produce a fake, the other trying to detect it. This competition continues over many cycles, resulting in a more plausible rendering of, for example, faces in video. GANs generally produce more convincing deepfakes but are more difficult to use. https://www.gao.gov/products/gao-20-379sp

Mar20 Machine learning Así funciona la tecnología que hace posibles los ‘deepfakes’ https://retina.elpais.com/retina/2020/03/12/innovacion/1584007649_777029.html

Como identificar

https://sempreupdate.com.br/5-maneiras-de-diferenciar-os-videos-deepfakes-da-realidade/

Here are the visual signs to look for that indicate a manipulated video. https://www.sdentertainer.com/lifestyle/how-to-spot-a-deepfake-video-on-the-internet/

O que fazer quando a detecção falha;

 https://www.brookings.edu/research/fighting-deepfakes-when-detection-fails/

ABR20 https://www.darkreading.com/endpoint/authentication/the-rise-of-deepfakes-and-what-that-means-for-identity-fraud/a/d-id/1337633


SOBRE O ZAO:

ZAO :https://www.calcalistech.com/ctech/articles/0,7340,L-3770491,00.html

ZAO: https://www.cnbc.com/2019/09/20/hao-li-perfectly-real-deepfakes-will-arrive-in-6-months-to-a-year.html

ZAO: https://apps.apple.com/cn/app/id1465199127 https://www.onyolo.co/zao-app-download-android-ios/

ZAO http://exameinformatica.sapo.pt/noticias/internet/2019-09-02-ZAO-app-de-deepfakes-chega-ao-topo-da-App-Store-chinesa