Google ha Vuelto! Lanzó GEMINI 4 Hydrogen al fin GRATIS y SIN Límites — Transcript
Full transcript
- 0:00Just because everyone thinks you’ve
- 0:01fallen behind doesn't mean you’ve lost
- 0:03the race.
- 0:05Google has just returned by launching
- 0:07its latest model, Gemini 4 Hydrogen. It
- 0:10performs similarly to the top models
- 0:12from its competitors, but with one clear
- 0:14difference. It is accessible to everyone
- 0:17for free. I’ve been testing it
- 0:19thoroughly this past week, and honestly
- 0:21, it’s the fastest AI that offers the
- 0:23best results.
- 0:29To prove it, we will put it to the test
- 0:31with use cases like generating any
- 0:33digital asset, 3D simulations, complex
- 0:35full-scale games, or even creating
- 0:37videos, among other use cases.
- 0:40The best part of all this is that beyond
- 0:42being a free model, even if they haven't
- 0:44publicized it, it could potentially be
- 0:46used unlimitedly. Sounds good? Well,
- 0:48let’s get to know all about it.
- 0:52Let’s start by getting to know a little
- 0:55better what this new Gemini 4 Hydrogen
- 0:56model consists of. The fact is, this is
- 0:59the replacement for the Gemini Flash
- 1:01models. Google reports that it is up to
- 1:042.5 times faster than its previous model
- 1:06, Gemini 3.8 Flash.
- 1:09It is better at programming, jumping
- 1:12from 18.2%to 57%on Terminal Bench 4.1.
- 1:16Artificial Analysis even positions its
- 1:18reasoning above the Argo model, which
- 1:20was only available as a paid service.
- 1:23It is a model that also contains a
- 1:24context window of over 1 million tokens
- 1:26, is multimodal—able to understand
- 1:28images, audio, video, documents, and
- 1:30text—and is optimized for all agentic
- 1:32and tool functionalities.
- 1:35But this is not just another model
- 1:37presented as an improvement on Google's
- 1:40line, as it is currently in the top five
- 1:42, above models like Faibol 5.1 or GPT
- 1:45Astra 6.
- 1:46Except this one is even faster and free
- 1:49, and its results are surprising.
- 1:52Take a look at this comparative Motion
- 1:54Graphics video where we have a superior
- 1:57model like the Gemini 4 Carbon and the
- 1:59new Gemini 4 Hydrogen, where we can see
- 2:01it did it in practically half the time
- 2:03and at less than half the cost.
- 2:06But what happens if we compare it to
- 2:07other models like Claude?
- 2:09Well, here with this animation of a
- 2:11floating house, we can see how Favol 5.2
- 2:14had a cost of about 19 dollars and took
- 2:16over 45 minutes to work, while if we go
- 2:19directly to this new free model from
- 2:21Google, it was the fastest by far,
- 2:23taking barely 12 minutes and costing
- 2:26approximately just 1 dollar.
- 2:29It looks very promising, so let’s
- 2:30discover where we can use it, and watch
- 2:32out, because its power changes a lot
- 2:34depending on where we are using it.
- 2:37We are going to start with the official
- 2:39Google Gemini platform, since now, even
- 2:41if you have a free version, like in my
- 2:43case with this account here which, by
- 2:45the way, will be the same free account
- 2:47we will be using throughout the entire
- 2:49tutorial, when we click on the model
- 2:51selection section, we no longer find the
- 2:533.8 Flash model, because even though we
- 2:55only see that we cannot use Gemini 4
- 2:57Argon, the truth is that by default all
- 2:59models are Gemini 4, meaning that now
- 3:01the free versions have those from
- 3:03Hydroge.
- 3:04From here we will also be able to switch
- 3:06between the level of reasoning, and this
- 3:08really changes the result a lot. Let's
- 3:10start by putting it to the test. For
- 3:12that, I will tell it here to simply
- 3:14create a professional website for an AI
- 3:16automation agency. I click send and
- 3:18notice that it starts programming right
- 3:21here, and if we go to the code section,
- 3:23we can see how all this code is really
- 3:25being generated very quickly.
- 3:28A little more than a minute later, we
- 3:30had the result here, and if I expand it
- 3:32, we can see how it has laid out this
- 3:34very futuristic webpage, fitting for
- 3:36that AI automation agency.
- 3:39From here we can see different blocks,
- 3:40and some calculators also appear that we
- 3:42can interact with, as you can see right
- 3:44here.
- 3:45We can see other sections, and if we go
- 3:47all the way to the bottom, we can see
- 3:49how it has actually completed the entire
- 3:51website, and most surprisingly, in
- 3:53record time. Let's put its physics to
- 3:56the test now.
- 3:57To do this, in a new chat, I will enter
- 3:59the following prompt asking for a 3D
- 4:01medieval siege game with physics.
- 4:03In short, a wall that appears so I can
- 4:05knock it down to see all the movements
- 4:07and how natural they are on the first
- 4:09try.
- 4:10To see this, I will click send again,
- 4:12and in just about a minute later, we
- 4:14have this simulator here.
- 4:16To see it better, I will expand it, and
- 4:18notice how we have this wall here; by
- 4:20zooming out a bit, we can see how we can
- 4:22launch a stone from here.
- 4:25So if I click fire, we can see we would
- 4:27have missed, so let's change the angle a
- 4:30bit lower, I'll add a heavier stone and
- 4:32reduce the power, and if I click fire,
- 4:34look now, it seems like it's going to
- 4:36crash, slow motion activates, and we can
- 4:39see how everything is actually working
- 4:41well with each of the movements.
- 4:44And all of this on the first try, but
- 4:46watch out because normally the effort
- 4:48level is at the minimum.
- 4:50Returning to where we left off, notice
- 4:52that if I start a new chat from here,
- 4:54when clicking on the reasoning level, we
- 4:55will normally find it at the fastest and
- 4:57most efficient setting, but by moving it
- 4:59to high, we will be able to get a
- 5:01somewhat better result.
- 5:03I'll give it the following prompt where
- 5:04I want it to create a 20-second vertical
- 5:06motion graphic style video that talks
- 5:07about fun facts about octopuses and why
- 5:09they have three hearts.
- 5:10If I leave the reasoning level at small
- 5:12here and click send, the first thing it
- 5:14will tell me is that if I want to
- 5:16generate it, I should switch to a higher
- 5:17version, but the truth is that this is
- 5:19something not necessary to be able to do
- 5:21it. Notice that if I now reply for it to
- 5:23do it in code or to do everything
- 5:25possible to make it happen and click
- 5:27send, then effectively, it would tell us
- 5:29here that it is already working on this
- 5:30video. A minute or so later, we can
- 5:32already find the video here. To see it
- 5:34better, I'm going to enlarge it. We have
- 5:36the video here now, so I'm going to
- 5:38click play to see if it appears
- 5:39correctly. Notice here that an error
- 5:41actually appears, so here we would have
- 5:43to iterate, and really, not only did it
- 5:45not give us the result on the first try
- 5:47, but it also did it with a small error
- 5:49.
- 5:50However, if I now switch the reasoning
- 5:52level to high in a new chat and give it
- 5:54the same prompt, notice that right away
- 5:55, if we click send, notice that on the
- 5:57first try, without having to urge it to
- 5:59make this video, it would immediately
- 6:01start working on it, and a little later
- 6:03, it would have delivered this other
- 6:05result on the first try.
- 6:07Let's watch those 20 seconds to see if
- 6:08it did it better.
- 6:10>> This animal has three hearts. Two of
- 6:14them pump blood directly to the gills to
- 6:16oxygenate it.
- 6:20And the third one pumps that blood to
- 6:22the rest of the body.
- 6:25When it swims, that heart stops beating
- 6:27. So
- 6:28>> And look, a result that is much better
- 6:30where we can get more out of this new
- 6:32Gemini model. With what we've just seen
- 6:34, we can create incredible content with
- 6:36artificial intelligence, but there is
- 6:38something that can completely change how
- 6:40we perceive it: sound.
- 6:42That is why I want to introduce you to
- 6:44Epidemic Sound, where we have more than
- 6:4655,000 songs and 250,000 sound effects.
- 6:49Furthermore, from here we can search for
- 6:51everything in a very easy way.
- 6:53If, for example, we start looking for
- 6:55music that fits the type of content we
- 6:56need, artificial intelligence will
- 6:58directly provide us with different
- 7:00recommendations that we can preview and
- 7:02download. We will also be able to do the
- 7:04same thing with sound effects.
- 7:05We just have to specify what we are
- 7:07looking for, for example, intensity in
- 7:09transitions or highlighting specific
- 7:10moments, and we would have all those
- 7:12sound effects right here to download and
- 7:14implement in our project.
- 7:16And there is something fundamental here
- 7:17that we must talk about: their license.
- 7:19A subscription allows us to use their
- 7:21catalog in our content and platforms
- 7:23without having to worry about copyright
- 7:25claims.
- 7:26In addition, videos published during an
- 7:28active subscription keep their license
- 7:29even after you cancel it.
- 7:31I’ll leave the link down in the
- 7:33description so you can discover Epidemic
- 7:35Sound, whom I also thank for sponsoring
- 7:37this video.
- 7:38Take advantage now because with Black
- 7:40Friday you will get a 50%discount on an
- 7:42annual plan with Epidemic Sound, getting
- 7:44the biggest discount of the year. And
- 7:46once we know this, let's continue with
- 7:47the next block.
- 7:52We are going to continue now by seeing
- 7:54how we can really get even more out of
- 7:56the new Gemini 4 Hydrogen model.
- 7:58Because although we have seen that we
- 8:00have three effort levels for Gemini, the
- 8:01truth is that we have an additional one
- 8:03that makes the difference.
- 8:04To discover all this, we have to jump to
- 8:07Google AI Studio, as this is the
- 8:08platform where we can use all the models
- 8:11in the Google ecosystem in a more
- 8:13unlocked way.
- 8:14In fact, notice that if we click on the
- 8:17top right, clicking on the latest models
- 8:19, we will find this latest one, Gemini 4
- 8:21Hydrogen, which we will be able to use,
- 8:23unlike its other two models, for free.
- 8:26I should warn you that this wouldn't be
- 8:28unlimited, just like with Gemini, but
- 8:30throughout the video we will see how we
- 8:33can use this model limitlessly.
- 8:35But going back to where we left off, I'm
- 8:37going to click on this model here and
- 8:38notice that for the reasoning level
- 8:40section, by default, we will have it in
- 8:42this case set to high, but we will also
- 8:44have a fourth level which is extra high
- 8:46.
- 8:47To put it to the test, I'm going to
- 8:48enter the exact same prompt we used for
- 8:50the octopus video in Gemini to see if
- 8:51there's any difference.
- 8:54To do this, I'll click on Run, and after
- 8:55about 3 minutes, we have all the HTML
- 8:57code right here.
- 8:59I'm going to click on the Preview
- 9:01section so we can open it, and we get a
- 9:03preview right here, but along with that
- 9:05, it shows we have Google voices
- 9:06connected, which should make it sound a
- 9:08bit more natural.
- 9:11A few more options also allow us to
- 9:13upload our own voice, but let's see if
- 9:14the video itself actually looks better.
- 9:17So, for that, I'm going to download this
- 9:20HTML code, open it to see it in a larger
- 9:22view, and see how it sounds and looks
- 9:25here.
- 9:25>> Did you know that this animal has three
- 9:27hearts?
- 9:29Two of them, the branchial hearts, pump
- 9:31blood to the gills to collect oxygen.
- 9:35And the third one pumps that oxygenated
- 9:36blood to the rest of the body.
- 9:40But when it swims, this last one stops
- 9:42beating. That's why swimming is so hard
- 9:44for them.
- 10:08>> settings, we will be able to use the
- 10:10Gemini 4 Hydrogen model, which we can
- 10:12use for free, and it will also offer us
- 10:15better results, since here we will have
- 10:17an even higher level of power.
- 10:20Notice that if we click on this section
- 10:22here, we'll have a max mode where we
- 10:24force this model to work for longer so
- 10:26that it offers us better results.
- 10:30To put this to the test directly, I'm
- 10:32going to do an example I've done in
- 10:33other models, which would be creating a
- 10:35game similar to GTA 6, that is, an
- 10:36open-world game, but with a somewhat
- 10:38simpler style.
- 10:41For this, in the prompt section, I would
- 10:43have entered this entire prompt here,
- 10:45where I'm telling it to create a
- 10:47playable 3D game inspired by open-world
- 10:49games, like a Minecraft or a GTA 6,
- 10:51where I want everything to appear:
- 10:53streets, cars, planes, so I can drive
- 10:56everything and it really lets me
- 10:57interact completely with this world.
- 11:01I'm going to click here on send so it
- 11:03starts working on this, and after about
- 11:05a little over 10 minutes, which is a
- 11:07long time for this model, let's see what
- 11:10it has managed to do for us.
- 11:12In fact, here I find this open map where
- 11:14I see a motorcycle appears, and from
- 11:17here I also see other people appearing,
- 11:19and the whole city. I also see different
- 11:22cars appearing here, for example a taxi
- 11:24, and if I click E, well, I can see that
- 11:26from here I can drive and it would all
- 11:28be working perfectly. Let's see if we
- 11:30can find, look, I'm just seeing a plane
- 11:31over here, let's see if we can interact
- 11:33with it.
- 11:34For that, I'm going to get out; it seems
- 11:36like all the buttons are working
- 11:37perfectly.
- 11:38I click E and from here, well, I could
- 11:40also be flying this plane.
- 11:43It is true that the movements might need
- 11:45a bit of iteration, but the truth is it
- 11:46seems to work quite well all things
- 11:48considered.
- 11:49Let's see if it also gave us anything
- 11:51else because I see something over there
- 11:52at the end. Okay, I was saying that I
- 11:55saw a port over here, so let's try to
- 11:57land this and from here we'll see if I
- 12:00can get on this boat, which I see is
- 12:03also working perfectly. Also, I can
- 12:05interact here.
- 12:07I also see a map where different little
- 12:09dots appear, which I understand are
- 12:11different missions, and ultimately, we
- 12:13have this simulation that has never had
- 12:15such a detailed result that works so
- 12:17well on the first try and in so little
- 12:19time, since usually these types of
- 12:21examples used to take around 30 or 40
- 12:24minutes.
- 12:25The Quicksol one took about 10 minutes
- 12:26and it did it much better than previous
- 12:28tests I was doing.
- 12:29In fact, notice that if we go back to
- 12:31the game, we could find other options
- 12:33like switching to night mode, where we
- 12:35could see everything perfectly applied.
- 12:37I also see different views like, for
- 12:39example, this first-person one where I'm
- 12:41going to try to get on a vehicle like
- 12:43this motorcycle, and look at how I could
- 12:45ride it from here.
- 12:47I even see many details like, for
- 12:49example, trucks or even cranes like this
- 12:51one here that I'm going to check by
- 12:52entering it and changing the view here
- 12:54to see if I can also operate this crane
- 12:56, and as you can see, I could also do
- 12:58that.
- 12:59And look at a result that was great, not
- 13:01only because the model is good, but
- 13:03because we also forced it to spend more
- 13:05time working in that max mode. In fact,
- 13:08while that part was being done, I tried
- 13:10to do it in Gemini too, and look here at
- 13:12how we have a much worse result where we
- 13:14really don't have any detail, no
- 13:15vehicles or people appear.
- 13:17I can kind of walk or levitate through
- 13:19this map here, but the truth is the
- 13:21result doesn't even come close to what
- 13:23we've done directly in Google AI Studio
- 13:26.
- 13:27But be careful, because this isn't just
- 13:29about us having used an exclusive
- 13:31application to create apps, since
- 13:32directly in Google AI Studio's chat mode
- 13:34I asked it for that same game, it
- 13:36created it in HTML, and if I open it
- 13:38here with that chat mode and a higher
- 13:40reasoning level, well, look here at how
- 13:42the result would actually be much better
- 13:44.
- 13:46I also find those people, I find cars
- 13:48that I will also be able to interact
- 13:50with right away, and all of this would
- 13:52be working perfectly, just like other
- 13:54means of transport such as, for example
- 13:56, that plane we are approaching right
- 13:58now.
- 14:00And if everything has worked this well
- 14:02and on the first try, well, from here
- 14:03I'll also be able to get on, and from
- 14:05here we could really see that it would
- 14:07also be working perfectly, since we
- 14:09could steer it as we can see here and we
- 14:11would be flying over this city.
- 14:17In this next part, we will see how we
- 14:19could actually extract this model from
- 14:21the Google ecosystem to be able to
- 14:22include it in any tool or harness we
- 14:24work with.
- 14:25Because yes, another thing they have
- 14:27released is the ability to use the
- 14:30Google Gemini API using this latest
- 14:32Gemini 4 Hydrogen model for free.
- 14:35To make this possible, I have returned
- 14:37to Google AI Studio, and from here, if
- 14:38we click on this key section to go to
- 14:40the API key, we will be able to find the
- 14:42API keys we have created or, conversely
- 14:44, if we didn't have any, by clicking on
- 14:46create API key, we will be able to
- 14:48generate one.
- 14:49But before taking this key and bringing
- 14:51the technology to another tool, notice
- 14:53that if we click on the rate limit
- 14:55section, we will be able to see from
- 14:57here the entire breakdown of models and
- 14:59how much we will be able to use them
- 15:00with their APIs for free.
- 15:03For example, here in the first one, we
- 15:05find Gemini 4 Hydrogen, from which they
- 15:07tell us we have an RPM, that is, a usage
- 15:09per minute of up to five requests.
- 15:12Then here we would have the input limit
- 15:14per minute in terms of tokens, which
- 15:15would be about 250,000, and how many
- 15:17requests we can make daily.
- 15:19In this case, we would have 20.
- 15:21This will actually change depending on
- 15:23the model, but keep in mind that this
- 15:25Gemini 4 has exactly the same limits as
- 15:27the small Gemini Flash models we had
- 15:29previously.
- 15:31Knowing this and returning to the API
- 15:33key, once we create an API key and click
- 15:35copy, we could now bring all this
- 15:37technology to any artificial
- 15:39intelligence tool, whether it's our own
- 15:41or, as in this case, I will be using a
- 15:43harness which will be DeepSeek's.
- 15:46Since this application here is
- 15:47open-source, it's quite similar to Open
- 15:50Claude, Hermes, or similar tools that
- 15:52allow us to connect any model we want to
- 15:54this technology. In fact, notice that if
- 15:57I click on the plus section, go into
- 15:59settings, and then head to the models
- 16:01section, if I click add a new provider
- 16:04from here and select the Google provider
- 16:06, by pasting the API key we just copied
- 16:08and clicking apply, it tells us that the
- 16:11models are now available. Notice that if
- 16:13I now click the model selector here and
- 16:16go to the Google provider, I can see all
- 16:18the models we have available.
- 16:20However, when I did this, by default,
- 16:23the Gemini 4 Hydrogen model didn't
- 16:25appear, because even though models like
- 16:27Gemini 4 Argon show up here, the truth
- 16:29is you have to pay to use them.
- 16:32If this happens to you too, there's a
- 16:34trick: if I go back to the settings
- 16:36section, return to models, and edit this
- 16:38provider with custom configuration,
- 16:40notice that when I click search for
- 16:42available models, that model we are
- 16:44talking about really doesn't appear.
- 16:48However, if I update it to show all
- 16:50these models here and add it manually by
- 16:52entering this ID, which I'll also leave
- 16:54below in the description—which is simply
- 16:56putting Gemini 4 Hydrogen—then it should
- 16:59work.
- 17:01From here, we would enter the name we
- 17:03want to appear when we select it; in
- 17:05this case, I'll put "Hydrogen Free," and
- 17:07if I click apply, I can now find this
- 17:09Google model I named "Hydrogen Free."
- 17:12And if I now say "Hello, can you hear me
- 17:14?"
- 17:15and click send, then about 2 seconds
- 17:17later it will tell us that it receives
- 17:20us correctly from the Deep Seek harness
- 17:22and that, effectively, it can work with
- 17:24us.
- 17:26With this, we can really take this free
- 17:28technology, which has daily limits, and
- 17:30use it for free in other places beyond
- 17:33the Google ecosystem, having a very
- 17:35powerful model that is also very fast.
- 17:38But, at the beginning of the video,
- 17:39besides saying it could be used for free
- 17:41, I also told you it was unlimited. To
- 17:43see this, let's move on to the next
- 17:44block.
- 17:48In this penultimate block, we'll see how
- 17:50we can manage to use this latest Gemini
- 17:524 model for free, but also in an
- 17:54unlimited way.
- 17:55There is something here that hasn’t even
- 17:58been officially stated, but Google's AI
- 18:00mode is actually already using this
- 18:02latest model that Gemini has released,
- 18:05only here we don't have any kind of
- 18:07limit.
- 18:08For this, I’ve gone back to Google, but
- 18:10if I now click on AI mode from here and
- 18:12in the prompt section I tell it that I
- 18:14want to visually understand how a home
- 18:16in Spain consumes energy with a bit more
- 18:18information, when I click send, it will
- 18:20start processing this, and in just a few
- 18:22seconds, not only will it be giving me
- 18:24this information, but I’ll also have
- 18:26this interactive visual to understand it
- 18:28much better.
- 18:30Notice how here I can turn on, for
- 18:32example, the heating and the dryer, so
- 18:34the meter would be increasing, and as I
- 18:36deactivate things, this meter would also
- 18:38go down.
- 18:39Additionally, I can tell it here how
- 18:41much on average each of these things
- 18:43would be using so that when I click turn
- 18:45on all, it shows me what the meter would
- 18:47read, what the estimated distribution of
- 18:49monthly expenses would be, and in short
- 18:51, the whole application to understand
- 18:53this easily.
- 18:55And besides that, we get much faster
- 18:57answers here thanks to having this
- 18:58latest Gemini 4 model.
- 19:01In fact, to verify this, I am going to
- 19:04click to start a new chat within
- 19:06Google's AI, and if I ask it here if it
- 19:09could confirm which model it is using,
- 19:11I’ll hit send, and it would tell me that
- 19:14it is indeed using Gemini 4 Hydrogen
- 19:16directly in this chat, which, unlike
- 19:19Gemini, doesn't have the limits we find
- 19:21in that section where we would have 5-
- 19:24hour or weekly limits, since inside
- 19:26Google's AI mode, at least publicly,
- 19:30there are no limits at all.
- 19:32And from here, the whole process of
- 19:34searching across the entire internet is
- 19:35also optimized.
- 19:37In fact, let’s do another use case, and
- 19:39for this, from a new chat, I’m going to
- 19:40enter the following prompt where I’ll
- 19:42ask it to compare what it would be like
- 19:44to live in Madrid, Barcelona, Valencia,
- 19:46Málaga, and Jaén for someone who works
- 19:47from home so that I can see what
- 19:49purchasing power I would need to live
- 19:51comfortably.
- 19:53I click send, and it will start
- 19:55processing so that, shortly after, not
- 19:57only will it give us an answer based on
- 19:59real internet information, but beyond
- 20:02all this data and interactive maps, we
- 20:04will be able to find an interactive
- 20:06panel here where, even if the map of
- 20:08Spain isn't correct, it would tell us
- 20:10roughly where each location is and the
- 20:13approximate budget we would need.
- 20:17For instance, if I change it here to say
- 20:18I'd like to live in a family apartment
- 20:20where I do need to use a vehicle daily,
- 20:22it would tell me that with those monthly
- 20:23euros, I couldn't live anywhere except
- 20:25Jaén.
- 20:26And, honestly, it would be a bit tight.
- 20:29If I now increase the budget, look at
- 20:31how other locations start turning green
- 20:33and the green bar appears a bit low,
- 20:34meaning I'm more comfortable in those
- 20:36locations.
- 20:38And this way, we can easily see all that
- 20:40kind of information that would take us
- 20:42quite a long time to research.
- 20:45Mind you, we also have to talk about the
- 20:47downsides, because even though it's
- 20:48completely unlimited, these models are
- 20:50the smallest and fastest ones for
- 20:51searching and finding information as
- 20:53quickly as possible.
- 20:54So, in Google's AI mode, we wouldn't be
- 20:56using the most advanced reasoning that
- 20:58provides better results for complex
- 21:00tasks.
- 21:01In fact, to see what this translates to
- 21:03, from a new chat, I asked it to create
- 21:05that interactive open-world simulator
- 21:07game we saw earlier. And while it's true
- 21:09that it did it in record time, look at
- 21:11how the results really don't come close
- 21:14to what we saw previously. For example,
- 21:16if I click here on "change vehicle,"
- 21:18this car would indeed be working.
- 21:20As we can see here, if I click on "
- 21:22change vehicle," the plane would be
- 21:23flying, but it's something very simple
- 21:25that isn't really comparable to the
- 21:27application we saw before in Google AI
- 21:29Studio.
- 21:33Let's finish with a bonus.
- 21:35Just a few weeks ago, Google launched a
- 21:37new platform that lets us create video
- 21:39games for free, and it has just been
- 21:41updated with the latest Gemini 1.5 Pro
- 21:43model.
- 21:45To see this, we would have to go to
- 21:47Playground, which is this platform here
- 21:49that we can access for free, but there
- 21:51is a slight catch. And that is the fact
- 21:53that it can only be used in the United
- 21:55States. In my case, I'm using it from
- 21:57Spain because I have a VPN that is
- 21:59currently simulating that I'm in New
- 22:01York.
- 22:02Regarding the VPN, keep in mind that
- 22:04even though they aren't sponsoring this
- 22:05video, I'll still leave the one I use in
- 22:07the description below, since with Black
- 22:08Friday now, there are some great prices
- 22:10, especially with a coupon I'll share
- 22:11below.
- 22:13Knowing this, look at how a bunch of
- 22:15games appear right here, like Street
- 22:18scale, adventure, car, Monopoly,
- 22:20Minecraft-style, and zombie games.
- 22:23Or one that really surprised me was this
- 22:25motocross one, where if I click on "
- 22:27play now," we'll see a game that someone
- 22:30made using artificial intelligence. Here
- 22:32, to see this, I'm going to click play,
- 22:34we'll hit freestyle, I'll go to
- 22:36customize garage, and look at how this
- 22:39bike appears here as if it were a
- 22:41completely full game where I can change
- 22:43the bike's color. I'm going to leave it
- 22:45in red. I can also change the rider and
- 22:47which engine I want to use.
- 22:49In this case, I'll use this four-stroke
- 22:50one, along with the tricks I can do and
- 22:52what the controls would be.
- 22:55And if I click on start session, look at
- 22:57how we'd be with this bike, which is
- 22:58also totally playable.
- 23:00In fact, let's try to do a trick.
- 23:02Look at how it's working perfectly here
- 23:04.
- 23:05And it really is surprising that this
- 23:07was coded by just anyone. Actually,
- 23:09doing it is very simple. For that, if I
- 23:11click back here and click on the create
- 23:13section, from here we can specify any
- 23:16game we want to create. To do that,
- 23:18continuing with the motorcycle example,
- 23:20I thought of creating a game I used to
- 23:21be very addicted to, which was the
- 23:23typical 2D motocross game that had
- 23:24different obstacles you had to pass. To
- 23:26do this, I would just enter that prompt
- 23:29with all these requirements. I'm going
- 23:31to tell it I want it in 2D graphics, for
- 23:32one person, although notice that we
- 23:34could also make multiplayer games.
- 23:36I click send, and after a few minutes
- 23:38and it having created all the things it
- 23:39was asking me about—I told it to
- 23:41generate everything for me—I'd already
- 23:42have this video game here that look at
- 23:44how I'm going to be able to play.
- 23:46And from here, we have a visualization
- 23:48where we can even change the different
- 23:50levels, and from here, we can see it
- 23:51would really be very similar to these
- 23:53types of games.
- 23:55This is a platform that could really
- 23:56support another video, so if you are
- 23:58interested in learning more about it in
- 23:59detail, let me know in the comments so I
- 24:01can make one about it. With this, Google
- 24:03seems to have once again democratized
- 24:05the use of artificial intelligence with
- 24:07its latest accessible model. And this
- 24:09shows us that, in many cases, to move
- 24:10forward, sometimes you have to go back
- 24:12to the beginning.
About this transcript
This page contains the full transcript of Google ha Vuelto! Lanzó GEMINI 4 Hydrogen al fin GRATIS y SIN Límites by Alejavi Rivera, generated from the public captions YouTube serves with the video. The transcript has 4,757 words across 704 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.