Hernan Moraldo
@hhm
Followers
1K
Following
9K
Media
85
Statuses
3K
Google DeepMind. Machine learning and creativity. Veo, Phenaki.
Joined December 2007
Proud to be part of the incredible talented team that delivered Veo 3! State of the art video and sound generation, including music, dialogue, and more. You can already use it in Gemini, Flow, and Cloud Vertex API.
Video, meet audio. 🎥🤝🔊. With Veo 3, our new state-of-the-art generative video model, you can add soundtracks to clips you make. Create talking characters, include sound effects, and more while developing videos in a range of cinematic styles. 🧵
2
0
23
Prompt: "Bear writing the solution to 2x-1=0. But only the solution!"
29
67
550
Muchas gracias a vos Jaime por el recuerdo! Fue una increíble sorpresa. Sí, esa letra y dibujos son definitivamente míos, y no han cambiado mucho desde entonces.
Amigo Hernán: ¡cuánto te agradezco aquellas palabras que conservo en el cofre del tesoro!. Si algún padre me lee: echad toda la leña a la lumbre de la curiosidad de vuestros hijos. ¡Que la vida no extinga ese fuego, esa ansia juvenil de querer saber más!.
2
48
328
More Veo 2 examples. Prompt: "A cat wearing a suit and a top hat, while driving a tractor. The tractor has lots of hay on top. Cinematic."
Veo 2 is able to:.▪️ Create videos at resolutions up to 4k.▪️ Understand camera controls in prompts, such as wide shot, POV and drone shots.▪️ Better recreate real-world physics and realistic human expression. In head-to-head comparisons of outputs by human raters, it was
7
17
261
Prompt: Snail riding a bicycle. The snail has a large moustache.
3
1
67
@RyanHouchin @QiaochuYuan The main thing is you never see the colors of the Milky Way with the naked eye. It looks like a dim cloud, but it's huge, and very impressive if you know what you are looking at. But most people living in areas with skies that clear don't ever realize what it is.
0
0
54
Así que gracias por publicar tus juegos también! :D. Me resulta muy familiar el nombre Arkaful, tenés algún screenshot?.
1
6
50
Por esos tiempos no tenía acceso a internet (que recuerde), así que el material de revistas como ésa me resultaba muy preciado. Pero sobre todo, encontrar personas de edad cercana a la mía que desarrollaban juegos me resultaba muy inspirador y me impulsaba a seguir aprendiendo.
1
6
48
So excited to share Veo! Incredible work by an amazing team!.
Introducing Veo: our most capable generative video model. 🎥. It can create high-quality, 1080p clips that can go beyond 60 seconds. From photorealism to surrealism and animation, it can tackle a range of cinematic styles. 🧵 #GoogleIO
2
2
34
Proud to see the release of Veo v2!!.
Today, we’re announcing Veo 2: our state-of-the-art video generation model which produces realistic, high-quality clips from text or image prompts. 🎥. We’re also releasing an improved version of our text-to-image model, Imagen 3 - available to use in ImageFX through
2
0
31
More Veo 2 examples. Prompt: "A snail wearing pants. The snail is riding a bicycle, and has a large moustache." #veo2. Following a comment by @thelokasiffers
1
3
27
Example video generated with Veo, prompt: . The camera follows an octopus flying through a city with very high skyscrapers. A sign says "Veo". Cinematic, 35mm film released in 2024.
1
2
17
@javierota @TEDxRdelaP Es lo mismo que pensé yo, pero me contuve hasta ahora. Le pasaron la pelota a la sociedad por un error propio.
0
0
12
@ID_AA_Carmack I had some fun with that a few years ago IIRC Sonic was also one of my test games.
1
0
13
@sarahookr Which version are you watching? . The book is a lot better then the TV series. And Tencent version of the show is much closer to the book (though it is more dense because of that).
4
0
12
This one seemed impossible to get until now.
0
1
10
Google AI just announced the project we've been working on for a while, with Leopold Haller, Nathan Martz, Stewart Miles, Hina Sakazaki!.
#ML techniques have demonstrated the potential to positively impact game development but often require specialized skills. Today we share a system that requires no ML expertise, works on many popular genres, and can train an ML policy in less than an hour→
2
2
10
1
0
9
So cool!!.
Time for a fully AI generated music video! . Using @GoogleDeepMind’s impressive Veo video from I/O + ElevenLabs Music and SFX models. 🔥.
1
0
6
Incredible short movie made with Veo 2.
These tools have come a long way these past six months. "Zorgop Knows All". Made with Google Veo 2
0
0
10
@danyscht @javierota PBI per capita, como todo promedio, es un indicador muy engañoso de la distribución. Si una persona muy rica se muda a un pueblo muy pobre, aumenta mucho el PBI per capita sin que disminuya la cantidad de pobres.
3
0
8
More Veo physics.
0
0
9
StoryBench is a new paper for movie generation / evaluation using Phenaki. Fantastic work by Emanuele Bugliarello and team!.
Wouldn’t it be cool if AI could help us generate movies?🎬.We built a new benchmark to measure progress in this direction🍿. “StoryBench: A Multifaceted Benchmark for Continuous Story Visualization”. 📄 👩💻 📈
0
1
8
Awesome news about our work on Veo!.
On @YouTube, we want creators to be able to express their creativity, build community + drive long-lasting businesses. New tools at #MadeOnYouTube are helping: we’re bringing Veo to Dream Screen to create high-quality, custom backgrounds on Shorts + more.
0
1
8
genial, y un poco espeluznante, la idea de 33 bits http://t.co/mzQ1hYkV lo que escribís con un nick puede ser relacionado con vos.
2
2
8
Alguien se dio cuenta que si traducís (mal) el nombre Lio Messi, te queda Lío Lioso en español, y Mess Messy en inglés?. Es quizás el primer pensamiento original sobre fútbol que tengo en mi vida.
1
2
8
Today we are glad to introduce Phenaki, a model that can generate videos from text, including multi prompt "stories". Check for details and examples!.
1/ Today we are excited to introduce Phenaki: short-link-to-paper, a model for generating videos from text, with prompts that can change over time, and that is able to generate videos that can be as long as multiple minutes!
1
1
6
Glad to see our work in Phenaki featured in the Google Research blog .
0
2
7
This thread has a lot of great Veo examples.
0
0
7
@joe_philleo @andy_matuschak Agree. Also I think there are other dimensions in which such writing isn't linear, which matter more. See for example
1
0
7
@sehurlburt Mathworld is typically better for that .Not sure whether it is too technical for this example still.
0
0
6
@Hana82784949 @MalaPersona10 Sí, era yo. Increíble leer esa carta nuevamente tanto tiempo después.
0
1
6
Imagen 3 also released today, incredible work by their team!!!.
We’ve also enhanced Imagen 3’s ability to:.▪️ Produce diverse art styles: realism, fantasy, portraiture and more.▪️ More faithfully turn prompts into accurate images.▪️ Generate brighter, more compositionally balanced visuals. →
0
0
5
I have a joke about overfirtting, but I would be the only one to laugh at it.
2
0
5
I love this image displaying the full observable universe (in log scale, by Pablo Carlos Budassi). Nice way to display planets, galaxies, and large scale structure in a single image.
2
0
5
@NotBrain4brain Just using the same VideoFX version people can use externally, nothing else. I didn't try other math problems, except some simple arithmetic. When I saw the arithmetic worked I moved to this formula, but didn't push further.
1
0
5
This is a great point!.
Video models still make lots of mistakes with physics, but they also get a lot right. My intuitive physics model is constantly getting feedback from the world to keep it on track yet we ask video models to produce entire clips with no sensory feedback built into the system.
0
0
5
AI Dungeon 2 is maybe the greatest innovation in games in long years. If you care at all about games you should try it out.
AI Dungeon 2 is publicly released! You can download it on iOS now and Android in another day or two. As a holiday gift we've decided that it will be completely free till the end of the year so try it out now!
0
0
5
@BertholdGambrel @paulg @ArnaudBonzom Also in some cases with multiple numbers: "ranging from 3 to 124" looks better than "ranging from three to 124".
0
0
5
@monicaihibert Te recomiendo eliminar esas preguntas, o transcribirlas. Es muy sencillo para alguien que reconoce que visitaste su escuela determinar quién es cada persona. Es ilegal publicarlas de este modo de acuerdo a por ejemplo ver 28.1.
1
0
4
@jasonbaldridge It's not exactly the same, but this reminds me of what people used to say about Niels Bohr:
0
2
4
@_johnwage @karpathy If you have concrete projects, as you make progress on them you find obstacles and gaps in your knowledge. You make progress by learning how to deal with those on demand, as they show up.
0
0
4
The approach described here is similar to what I presented in 2014 in (I think others adapted it to neural networks in the subsequent years, but the basic idea is easily generalizable to any random process).
OpenAI is working on a watermarking tool to mark GPT-generated text with a special, hidden signal. And they have a working prototype.👀. How does it work?👇🏻.
1
0
4
Great summary of our project!.
Honored to have @sundarpichai tweet about our work!! 🥳🥳 Checkout our astronauts, pandas washing dishes, driving, riding bikes, and other cool videos in our website #Phenaki #text2video.
0
0
4
Yes, the idea of improving LLMs incrementally by training on their own output after using chain of thought is very interesting!.
@I_are oh, I'm pretty hyped up about this one. One of rare occasions I used an !!! in my newsletter.
0
0
4
A literature teacher once told my class that "million" or "millions" was almost always exaggeration. She said there is just not that much stuff around us on a daily basis, and advised us to use hundreds, and sometimes, thousands. Now in tech, millions is often too little!.
0
1
4
6 years ago I left Argentina to join Google (first in UK, later in the USA). Crazy how the time flies. and it's been a ride!.
0
0
4
Now I'm a dad I can finally say my bad jokes are dad jokes. I've been training all this time!.
0
0
4
@ID_AA_Carmack The workplace notes really look like your old .plan files (maybe more so than tweets).
1
0
2
Beginners are also incredibly useful for finding important gaps in documentation, processes, etc. and creative solutions to known problems. The beginner state is precious, never to be wasted.
@sehurlburt From another perspective, beginners are often in the best position to teach(write) about the things they just learned. They do not make as many assumptions of previous knowledge as many experienced people do, and thus it often helps it "click" for other beginners.
1
0
2
@D_Twitt3r @paulg @steven_braham @MegaBasedChad Only caveat is that it will fall back to the default every now and then, and when that happens you have to set it again to latest tweets.
0
0
3
@Lijter @BrunacciHernan @AlejandrOK2012 El libro se escribió mientras se hacía la película, y se publicó después.
0
0
3
@kevin_zakka @hardmaru I suspect the reason why we are still benefitting from a form of "feature engineering" in IL / RL is that agent tasks are a lot more sensitive to errors (b/c of how error propagates through steps in an episode) than Supervised Learning tasks.
1
0
2
@michael_nielsen It's interesting how different this is from Richard Hamming's comment on working in the most important problems in your field.
1
0
2
@squareoctopus @matajueguitos @__rumpel Ah, me dijo ya lo contactaron por la entrevista así que no tengo datos muy misteriosos para aportar. Hablábamos un montón en esa época en que estaba trabajando en sus juegos, y me parece genial que se recupere y valore esa parte de la historia.
0
0
2
Wow! So proud to have helped with its foundation, even if I haven't been around to see it grow for at least a decade. Great work and congratulations!.
ADVA, the Argentinian Game Developers Association, turns 16 today. I'm extremely proud of being a part of it since it's origins. ADVA shaped the industry in our country and made grow it into what it is now by attracting talent and helping studios. We need to continue growing!.
0
0
3
Very cool video generated using Phenaki and Imagen Video.
0
1
3
@JenovaChen My 5.5 y.o. wanted to send you this letter. She's been playing Journey since she's 3 and a half! A big fan, she draws the traveler everywhere, makes stories with her younger sister, and plays with dolls. Thanks to you and the team for bringing so much joy to us!
1
0
3
It's always great to return to the Computer History Museum, in Silicon Valley. The first time I went there it was an incredible experience, and I remember feeling echoes of what I felt in my first trip to the US, the day I went to the MIT Museum.
2
0
3
Another cool example of a video generated from text, using Phenaki and Imagen Video.
AI short film created with text-to-video. At the rate of advance in this area of research, we are likely ~2 years out from seeing a major tv show fully created using similar techniques. prompt in comments!. #generativeart #AIart
0
0
3
@JeffDean @OriolVinyalsML The chart was very nice! It used to be visible to the public in the old Visitor Center (beta). I don't know where it is now.
0
0
3