<section data-background-iframe="https://www.youtube.com/embed/2fAPgOCjToA"> <div class="title-text"> <h2>Generare: Immagini</h2> Gabriele Graffieti Forlì, 13 Febbraio 2024 <!-- .element: class="smaller grey italic" --> </div> Note: Interacion and questions </section> --- ## Chi sono <div class="two-c-container" data-markdown> <div class="two-c-col-l"> <img src="./img/mine.jpg" width="90%"> </div> <div class="two-c-col-r" data-markdown> - Sr. Algorithm Engineer @ [Ambrella](https://www.ambarella.com) - Deep Learning & Computer Vision - Head of AI research @ [AI for People](https://www.aiforpeople.org) - In passato ricercatore \& PhD @ [Unibo](https://www.unibo.it/en/homepage) - Interessi di ricerca: Continual Learning, GenAI and Ethics </div> </div> --- <!-- .slide: data-auto-animate --> ## Perché usare o generare immagini? --- <!-- .slide: data-auto-animate --> ## Perché usare o generare immagini? - Perché no? - Una immagine vale più di mille parole <!-- .element: class="fragment" data-fragment-index="1" --> - Nessuna differenza di lingua, cultura, interpretazione <!-- .element: class="fragment" data-fragment-index="2" --> - Umani e maggior parte degli animali hanno la vista come senso più sviluppato <!-- .element: class="fragment" data-fragment-index="3" --> - Unico modo in molti ambiti applicativi <!-- .element: class="fragment" data-fragment-index="4" --> --- ## Esempi: - Auto a guida autonoma <!-- .element: class="fragment" data-fragment-index="1" --> - Sistemi di sorveglianza e controllo <!-- .element: class="fragment" data-fragment-index="2" --> - Medicina per immagini <!-- .element: class="fragment" data-fragment-index="3" --> - Analisi immagini/video <!-- .element: class="fragment" data-fragment-index="4" --> - Fotoritocco <!-- .element: class="fragment" data-fragment-index="5" --> - Generazione di immagini/video/arte/effetti speciali <!-- .element: class="fragment" data-fragment-index="6" --> --- <iframe width="1120" height="630" VQ=hd1080 frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen data-src="https://www.youtube.com/embed/xhZfGCwU-GY"></iframe> --- <img src="./img/Overreaching.gif"> --- <img src="./img/MachineAreaControl.gif"> --- <img src="./img/incidenti.gif"> --- <iframe width="1120" height="630" VQ=hd1080 frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen data-src="https://www.youtube.com/embed/YHTSdd8-bnc"></iframe> --- <img src="./img/photoshop.gif"> --- <img src="./img/pope.png" height="630"> --- <img src="./img/mj1.jpg"> Falsa <!-- .element: class="fragment custom blur" --> --- <img src="./img/mj2.jpg"> Falsa <!-- .element: class="fragment custom blur" --> --- <img src="./img/vera.jpg"> Reale <!-- .element: class="fragment custom blur" --> --- <img src="./img/mj3.jpg"> Falsa <!-- .element: class="fragment custom blur" --> --- <img src="./img/mj4.jpg"> Falsa <!-- .element: class="fragment custom blur" --> --- <img src="./img/mj5.jpg" height="550"> Falsa <!-- .element: class="fragment custom blur" --> --- <img src="./img/AiArt.png" height="630"> --- <img src="./img/debellamy.avif"> Edmond de Belamy <!-- .element: class="smaller grey italic" --> --- <!-- .slide: data-auto-animate --> <img src="./img/klingermann.jpeg"> Mario Klingemann <!-- .element: class="smaller grey italic" --> --- <section> <h2>LLM in 5 minuti</h2> </section> <section> <img src="img/transformer.png" height="630"> </section> <section> <img src="./img/mllm.gif"> </section> <section> <img src="./img/longtail.png"> </section> --- ## Analisi di immagini/video - <!-- .element: class="fragment" data-fragment-index="1" --> Semnatic e Contextual Search su video (<a href="https://search-oyntfgdwsq-uc.a.run.app/">demo</a>) - <!-- .element: class="fragment" data-fragment-index="2" --> Comprensione di immagini complesse (<a href="https://karpathy.github.io/2012/10/22/state-of-computer-vision/">blog Andrej Karpathy</a>) --- <!-- .slide: data-auto-animate --> <img src="./img/obamafunny.jpg"> --- <!-- .slide: data-auto-animate --> <div class="two-c-container" data-markdown> <div class="two-c-col-l"> <img src="./img/obamafunny.jpg"> </div> <div class="two-c-col-r" data-markdown> <img src="./img/obama_gpt1.png"> <img src="./img/obama_gpt2.png"> <!-- .element: class="fragment" data-fragment-index="1" --> </div> </div> --- ## Generazione di immagini/video - <!-- .element: class="fragment" data-fragment-index="1" --> Generazione di immagini tramite prompt testuali (<a href="https://chatgpt.com/">dall-e</a>, <a href="https://www.bing.com/images/create">bing</a>, <a href="https://copilot.microsoft.com/">copilot</a>) - <!-- .element: class="fragment" data-fragment-index="2" --> Generative Photoshop (<a href="https://www.canva.com/features/ai-photo-editing/">canva</a>) - <!-- .element: class="fragment" data-fragment-index="3" --> Generazione personaggi digitali (<a href="https://runwayml.com/research/introducing-act-one">runway</a>) - <!-- .element: class="fragment" data-fragment-index="4" --> Effetti speciali (<a href="https://www.youtube.com/watch?v=sK991q0mTBA&t=256s">pika ai</a>) --- ## Rischi ed etica - Deepfakes <!-- .element: class="fragment" data-fragment-index="1" --> --- <div class="two-c-container" data-markdown> <div class="two-c-col-l"> <iframe width="1120" height="630" VQ=hd1080 frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen data-src="https://youtube.com/embed/ErXgb4cAVIM"></iframe> </div> <div class="two-c-col-l" data-markdown> <iframe width="1120" height="630" VQ=hd1080 frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen data-src="https://youtube.com/embed/RJwwzJosfN4"></iframe> </div> </div> --- <img src="./img/kamala.png" height="500"> [link](https://x.com/elonmusk/status/1816974609637417112?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1816974609637417112%7Ctwgr%5E8d0f7a529f12f49248f45d162ddc3039761257fe%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fiframe.nbcnews.com%2FlQLMZqL%3F_showcaption%3Dtrueapp%3D1) --- <img src="./img/kamala2.png"> [fonte NBC](https://www.nbcnews.com/tech/misinformation/kamala-harris-deepfake-shared-musk-sparks-free-speech-debate-rcna164119) --- ## Rischi ed etica - Deepfakes - Fake influencer <!-- .element: class="fragment" data-fragment-index="1" --> --- <!-- .slide: data-auto-animate --> <div class="two-c-container" data-markdown> <div class="two-c-col-l"> <img src="./img/fakeInfluencer.webp"> </div> <div class="two-c-col-l" data-markdown> <img src="./img/dulceida.png"> </div> </div> --- <!-- .slide: data-auto-animate --> <img src="./img/fakeInfluencer.webp"> --- <img src="./img/fakeAiInfluencer.png"> --- ## Rischi ed etica - Deepfakes - Fake influencer ([Aitana Lopez](https://www.instagram.com/fit_aitana/?hl=en), [Lil Miquela](https://www.instagram.com/lilmiquela/?hl=en)) - Innamoramenti di persone finte <!-- .element: class="fragment" data-fragment-index="1" --> --- <img src="./img/innamoraChatbot.png"> <img src="./img/innamoratoChatbot.png"> <!-- .element: class="fragment" data-fragment-index="1" --> <a href="https://www.leggo.it/esteri/news/donna_innamorata_chatbot_intelligenza_artificiale_marito_benessere_mentale_oggi_21_1_2025-8607473.html">Fonte 1</a> <!-- .element: class="fragment" data-fragment-index="2" --> <a href="https://www.corriere.it/tecnologia/24_ottobre_23/innamorato-di-un-chatbot-un-adolescente-americano-si-e-suicidato-a-14-anni-la-madre-fa-causa-all-app-pericolosa-e-non-testata-45697ab8-7316-46d3-bc01-60de5bb74xlk.shtml">Fonte 2</a> <!-- .element: class="fragment" data-fragment-index="2" --> --- ## Bonus - FantaSanremo <!-- .slide: data-auto-animate --> <div class="r-stack"> <img src="./img/fantasanremo.png" height="300px"> <img class="fragment fade-out long-transition" src="./img/fantasanremo2.png" height="300px"> </div> --- ## Bonus - FantaSanremo <!-- .slide: data-auto-animate --> <div class="two-c-container" data-markdown> <div class="two-c-col-l"> <img src="./img/fantasanremo.png" height="300px"> </div> <div class="two-c-col-r" data-markdown> Ad oggi (13/02) <b>2/19</b> con 660 punti. </div> </div> --- <img src="./img/thatsall_white.png" height="300px"> --- <section data-background-iframe="https://www.youtube.com/embed/2fAPgOCjToA"> <div class="title-text"> <h2>Domande?</h2> </div> </section>