--- title: 'Warning: Gemini Omni Is Amazing, But There''s Something You Need to Know' source: 'https://youtube.com/watch?v=teG2L1LOAkA' video_id: 'teG2L1LOAkA' date: 2026-07-23 duration_sec: 899 channel: 'Rob Boliver - Como usar o CANVA' --- # Warning: Gemini Omni Is Amazing, But There's Something You Need to Know > Source: [Warning: Gemini Omni Is Amazing, But There's Something You Need to Know](https://youtube.com/watch?v=teG2L1LOAkA) ## Summary Google's Gemini Omni is a powerful new video generation model that can create realistic videos from photos and prompts. However, the model is not yet fully ready, and users should be aware of its limitations, including high credit costs, inconsistent results, and regional restrictions on features like live editing. ### Key Points - **Gemini Omni Overview** [00:13] — Google released Gemini Omni, a new video creation model that generates videos from photos and prompts. It works in Google Flow and Gemini, but only for paid plans above Pro. - **Model Not Fully Ready** [01:07] — Google stated during the live stream that the model is not 100% ready yet. It is currently the Flash model, a faster but smaller version, with a full model coming later. - **Video Editing Capability** [02:57] — Users can upload their own video and add effects or edit by talking. However, this feature is not available in Europe due to legislation. - **Comparison with Sora** [03:50] — Gemini Omni is compared to Sora, but the creator believes they are different models. Omni allows editing, while Sora is more advanced in generation. - **Example Creations** [05:26] — The creator tested prompts for historical scenes and commercials. Results were mixed: some realistic, others with errors like distorted faces or physics issues. - **Credit Costs** [07:26] — Each creation costs 30 credits. The creator's Ultra account had 10,000 credits, but another account with fewer credits was quickly depleted. - **Live Editing Feature** [08:07] — A key differentiator is the ability to edit videos live by speaking. This feature is not yet available in Europe. - **Geographic Accuracy Issues** [09:53] — The model sometimes misrepresents landmarks, e.g., Maringá Cathedral was incorrectly generated, showing the model's limitations in accuracy. - **Cost Comparison with Sora** [10:44] — Gemini Omni is much cheaper than Sora, but multiple attempts are often needed for good results. A film at Cannes using Sora cost hundreds of thousands of dollars. - **Commercial Use Recommendation** [12:33] — The creator advises using more established models for commercial purposes until Gemini Omni improves, as it still requires many credits and attempts. ### Conclusion Gemini Omni shows great potential for video creation and editing, but it is not yet reliable for commercial use due to inconsistencies and high credit costs. Users should test it for learning but wait for a more stable version for production. ## Transcript template that Google just released. I used a photo and a prompt, and it generated this super cool video. But hold on, before you go ahead and spend your new model that promises to be one of the best on the market, you need to watch this video here. Because since its release, I've made a video here on the channel seen it yet, I'll leave a link here so you can watch it later. Many tests were performed. I myself have already spent thousands of my Google credits here to test and create, to show you. And you might change your mind after want to be left behind, watch until the end. I'm Rob. Here we talk artificial intelligence, and tools that can help you be much more productive and make type of content, subscribe here. We have videos like this every week, okay? Look, for those who don't know what I 'm talking about, and just stumbled upon this, Google launched Gemini Omni, which is Google's new, super-smart video creation model. But the model isn't 100% ready yet. Google itself made this very clear during the live stream they did. There was were talking about this. So you can see here that I've already used up a lot In this account here I have several examples to show you. And I've also selected some examples of what people are creating with this enormous potential, but it also has a little problem that you need to know about. Well, the prompt I used to make that video at the beginning was found it on Twitter, and I thought it was super cool, so I downloaded it and made a version of the prompt here so I could create it. I also made a version with Brazilian cities. What 's the point of the prompt here? It's a super simple prompt. I went inside Google this new Google model works in both Google Flow and Gemini. So you can access it through Gemi, but only for paid plans above the Pro plan. Okay? So, before you invest your money there, Google will also release this feature for creating short videos on YouTube. So, just imagine what's coming next, okay? So, I took that prompt, adapted it, and it turned out pretty cool. This prompt here almost went viral over there on my ex. I found it very interesting, very creative. And it perfectly demonstrates the model's ability to create super-realistic images based on the you can see that he brings the images here, and I've been to several of these places, and he really does bring them here, look, here 's Big Ben in London, right? He brings them here, look, here's the opera house in Sydney, Australia, and he brings the images with great precision . But before you get to a try two or three times to get a video that's actually good. But this happens with all models, okay? One really cool thing about this new Google video model is that you can upload your own video and ask them to add an effect, editing the video just by talking. I tried doing this in several different ways. I know there are some features that aren't available here in Europe yet. That's usually how it is. Some first due to legislation and then they come here. So, I believe blocked precisely for that reason. I tried to edit my own video there and could n't at all, even following the tips from find anything specific about it, but I believe that's it, I work think that's what's happening. So, in this case, the guy put a prompt there, like it was a ball of power, and he's blowing on it. The result is really cool, but obviously many creations must have been made to arrive at this result here. This was a really interesting video, comparison between Sedents, which is the most advanced video creation model today . I've even made some videos with Sidens, I posted them on my other social media, you can check them out later , okay? Comparing it to the OMN. So, a lot of people think that Omni came to compete with CEN. I don't think that's it, okay? I think they are different models, and the OM allows you to edit them when they release the full model because this is still the Flash model, that is, the faster model , but let's say a little smaller than the model that is coming. Google itself made that clear. So they release these models for us to test and for it to evolve, building the they go ahead, release it, people test it, they improve the model, and that helps bring in more information to improve the next model. So that's what to improve the next model. So that's what they're going to do. Look at this, realistic, man. Very good. Very good. And here, look how it is here. Google seems like [groan] their model didn't turn out well, did it? Hey, take a look here. Oh, that take a look here. Oh, that comparison! Shadows. Google, on the other hand, seems to have gone into slow motion and all that. And I want to show you some examples of mine as well, so you can see. So, I'm going to show you what I've done, what I think has potential, and what perhaps doesn't. So, let's go . Look here, I asked for it to be done as if it were from the past, right? So, I'm using Flow here, okay? You can use it in Gen and here, look , I asked him, following all the parameters of Google itself, how you can create your prompts. So, I took that manual, put it exactly as Google asked. Look at this, man. I'm here in Rome. Rome itself. Look at this , aren't they? Which one is this? You can see that people are kind of closed off here. He's already messed up here, right? And hey, for those who are ultra, there's no digital watermark, so you can if you bring it to Google. Google even talked about it, but there's no watermark here. So this is something they will most likely do when they release the API, that is, the connection key that you can use to create your start using it, there will be no watermark, which is good news. good? Response. See? Which one is good? I understood. I understood. It's not bad at all, you know? Hey guys. All good? Shy, this ending here was kind of strange. Here's another take. Take a man. I'm here in Rome. Rome itself. That's really cool. I thought this part was really cool, didn't it? Very realistic indeed. Look at this. People are kind of good. I understood. I understood. See? I started talking and the model got confused. So, the other person started talking as if it were me talking, kind of imitating me, you know? It 's not bad at all, you know? Hey guys. All That's why I asked to do a commercial. Just look. I found a photo of my make a commercial out of it. Want to see? Let me super simple prompt. Just look. I uploaded my photo. He's doing a professional advertisement and I've included a picture of the camper here. He put it here. [laughs] The campervan got wrecked. my voice here yet. As I said, this option hasn't been approved here in Europe yet . For those who asked in the last video about the credits, each creation is worth 30 credits here, okay? So it's quite a lot, depending on your account. I have the Ultra account and I've already spent a lot, a lot, right here on my account. I think I have 10,000 credits there to use. But I also have another account, an account that I use for work, and that account, well, it's gone now. I used up all my credits there. There, I think I had, I don't know , maybe 1 or so credits. Meet Meet was good. live that it's like this, I can come here and say the following, look. He's speaking Brazilian Portuguese, OK? Then you can request to edit it live here. This is one of the major differentiating factors of this model compared to other models. So you can do the editing here very quickly, right? That's why they're calling it the nano banana of videos. And here are a few more examples. example. The person went there and asked, you know, to be able to zoom in here on the Big Bang. Very good. He's very good with this localization issue and all that, but with physics in really adhering to prompts and such. And here's a really cool example. Just look. Looking into the eye of the Mona Lisa. Very good guy. He's also very good at creating explainer videos. Look how fantastic! And you don't need to use super complex prompts to be the future will be as follows. This template will be very good for people who want to use it for editing. So I want to put something here in the background, I want to change something, I want to create an effect, like someone who doing that just by editing like that. I want to come back here and show you some more examples that frustrated me a bit, okay? So you'll see here, in this case, I took the same prompt as the woman and just asked the cloud to Brazilian cities. He did it, it turned out well, but the way he used it, there were about three different ways the girl could be positioned there, yeah, there are some things here that didn't turn out very well, but the knowing that video technology today can do this kind of thing is already really cool. [music] you know it's really good, right? Except here, look, for the Maringá Cathedral that he built here, look. I lived in Maringá. Maringá Cathedral is not like that. It's like a cone, but it's not. This is not the Maringá Cathedral. So he made a mistake here, look . The historic city there was one of the places I've visited before. The historic city of Paraty looks very similar. Iguazu Falls as well. Here you have the National Congress and Christ the Redeemer. It turned out great. The video turned out really cool, right? Yes, there's still room for trying to improve the video. Look, I asked again here. She put her hair there. It turned out well. I actually liked it, actually. It turned out great, much cheaper than the Sedens, which is the model that people are comparing it to a here and many of them didn't turn out well. So you can see that often the person artificial intelligence tries several times works. So what's happening more and more is that these models are becoming so good that there will come a time when you'll try three or four times and But for those who work with this, there's even a platform, if I'm not mistaken, it's called Hickfield. And they created an entire video, a whole film, to show at the Cannes Film Festival with Sedens 2.0. And they spent millions of credits, right? It cost hundreds of thousands of dollars, but if they were to pay to create that video in a movie theater, it would cost millions and millions of dollars. They about it. That's really cool. So, here I tried to make it like a vlog where I was there with the dinosaurs. Take a look, man. Dude, wait. I'm seeing what I think I'm seeing. My God. My God in heaven. Are you makes no noise. It makes no noise. For God's sake. Run, man. And what did I say? He mentions running at the end, right? Run. He screams, run! Then he ran. It did n't turn out so well. And look, I ordered another one. If I were vlogging in ancient Rome and wanted to pay with Pix. It's something funnier over there. When I put, man, a fast food place in Ancient Rome. Surreal. Deal? Oh, accept it, accept it. I'm waiting for the signal. It gets kind of confusing, doesn't it? Including what was said there. So, from time to time the model makes mistakes here too; in this case, when you enter the flow, So, you click here, the agent helps you improve your video, to create something really smart here, but then it causes problems. One thing I found very interesting is precisely the issue of advertising. So today you can create good ads with this tool, with just a very simple prompt, and also videos with super simple prompts, you can achieve commercial, you know, to see how it would turn out. Some turned out well, others turned out awful. Oh, this one I requested, I put up a picture of my camper and vehicle. This one here has already become quite interesting, but he modified interesting, but he modified the vehicle a bit. But look at the talent, man, . This is really good. In Ireland, that kind of location [music] really exists, but here it has already changed. This place is called Cliffs of Morer. If He's already moved here. So, he started in one location and moved. So it did model isn't 100% yet. So, if you 're thinking about going there and investing your money to start using this commercially, to learn how to do it, then yes, it's worth it, right? So you can understand do the tests, right? Each model works in a different way, I think that's very valid, but if you want to do this to use it commercially, you might end up using a lot of credits to get a good model, right, use the model we already know, which is Vio, which is the it also won't make those scene changes, those things that this model . Oh, this one is interesting. You'll see what happens. Look You'll see what happens. Look how crazy this is. Unashenture [music] went backwards, then it came back. really does have those [musical] references. It's clear that it has great potential. It's a cool model, but particularly like this, I invested with generative artificial intelligence, I'm constantly learning more about models, so I always need to be updating myself. But for those of you who might commercial way, I recommend that you use a more established model for now, until this one becomes better. And things are few weeks we'll already have a much newer and much improved version of this model. Comment below what you thought, if you've already think of these new features that Google has brought. There are so many that I couldn't even bring them all here on the channel, but I'm sure there's still a lot more video here for you, if you haven't watched it yet, so you can go and watch it. watched it yet, so you can go and watch it. See you next time. See you later.