[00:02] days ago and it's breaking the internet. This model can generate precise text inside images, cinematic shots, and details I've never seen from AI before. But the real question is whether it's actually that good or not. So today I'm
[00:14] putting it head-to-head with the current king of image generation, Nano Banana Pro. I'll test both of them using the same exact prompts across text rendering this video, you'll know exactly which one is actually worth using. So the
[00:27] rendering. And what I mean by that is how good these models actually are at creating precise text inside detailed images. The reason I'm starting with this is because most image generators completely fail at it. They might create
[00:40] a good-looking image, but the moment you ask them to add full paragraphs of text, everything starts to fall apart. And up until now, Nano Banana Pro was one of the few models that was actually decent at this. So for this comparison, I'm
[00:52] going to use OpenArt because it gives me access to both GPT image 2 and Nano Banana Pro in the same place. This makes the entire test way easier and faster different platforms or pay for multiple subscriptions. So if you want to follow
[01:06] link to OpenArt in the description below. Once you're inside, go to the image section and choose the image model. For the first test, I'm going to we even write the prompt, let's set up the settings that will stay exactly the
[01:19] same for both models. For the aspect ratio, I'll go with 16 by 9 and then available. Now we can move on to the prompt. For this test, the image we're generating is an anatomical rocket cutaway poster packed with labeled parts
[01:32] the prompt and hit generate. And here's what Nano Banana Pro gives back. At first glance, it honestly looks pretty solid. The layout is clean, the panels are organized well, and the typography is large enough to scan quickly. So if
[01:45] infographic, this would already be a pretty decent result. But once you actually start zooming in, that's where things begin to fall apart. If we look reference is misspelled as refference. And that's exactly the kind of mistake
[01:59] that instantly breaks the credibility of a poster like this. Then on top of that, the title is split into two lines with extreme on one line and tectonics right below it. Prompt specifically asked for one title in the top left corner. So,
[02:11] that's already a miss in terms of prompt accuracy. And the rocket itself is much asked [music] for a deep cutaway with But, what we get looks more like a polished diagram than a true technical
[02:24] cutaway. So, overall, the looks good on the surface. But, the moment you start it falls short. Now, let's run the exact same prompt with GPT image 2. So, I'll and the prompt the same, and hit generate. And the very first thing that
[02:38] stands out is the text. The title is one clean line, exactly like the prompt asked for. The footer is spelled correctly, there are no typos, and the systems legend, the specs table, and even the small captions are all readable
[02:50] and much more faithful to the prompt. But, it's not only the text that looks details. You can clearly see the inside parts, the tanks are easier to more like a real rocket instead of a simple drawing. The material rendering
[03:04] is also on another level. The copper and steel details, together with the warm paper texture, make it feel much closer to that vintage NASA poster style. And even these smaller supporting panels look better. They feel like real mini
[03:16] technical illustrations with more depth in the capsule interior and the engine than what we've seen from Nano Banana Probe. This was only the text rendering test. more category that's really important for both images and AI videos.
[03:30] And that's the cinematic look. So, basically, we care about how well these models can create atmosphere, depth, and lighting that make an image feel like a this matters so much [music] is because you can't really create high-quality AI
[03:43] videos without starting from a cinematic image. If you try to generate a video from a low-quality reference image, you'll struggle to get consistent and model you use. But if you have a cinematic one from the start, you're
[03:55] basically 80% of the way there. And I'm actually going to show you exactly that get there, let's see which of these models is better at creating that model back to Nano Banana Pro, paste in the new prompt, and hit generate. For
[04:09] astronaut with a massive planet in the background. And this is what Nano Banana Pro gives us back. The atmosphere, honestly, does look pretty cinematic right away. The dust haze and soft lighting give it that movie still feel.
[04:22] clean. But the more I look at it, the more it feels like something is off. For behind that haze and glow. And the astronaut, [music] especially, gets lost in the frame because the bright background is taking too much attention.
[04:35] Also, the planet itself has way less detail than I wanted. It looks nice from like a sharp, detailed planet in the scene. Now, let's run the exact same prompt with GPT image 2. And right away, the difference is noticeable. The first
[04:48] astronaut looks small compared to the planet, which makes the whole scene feel much bigger. The ground is also way more detailed. You can clearly see the red leading up to the astronaut. So the
[05:01] just looking like a soft background. The image is also easier to understand. Your astronaut and then to the planet in the background without feeling messy. So there's no doubt that GPT image 2 completely beats Nano Banana Pro at
[05:15] this, which means you now have the power to create high-quality cinematic images. Images that respect all the text details you give them, all in the same tool. But image can actually do. Because there's a feature inside OpenArt that uses it to
[05:29] create high-quality and consistent AI videos faster than pretty much any other workflow out there. It's called Smart Shot. And what makes Smart Shot different is what's actually inside it. The image side is powered by GPT image,
[05:41] the same model we just tested. And the video side is powered by Seed 2, which is currently one of the best AI video models out there. So, when you describe your scene, GPT image plans out the entire visual side from the character to
[05:53] then Seed ends turns that plan into the actual video, making sure all the motion exactly how this works. I'll head over to Smart Shot inside OpenArt. The first
[06:05] description of the scene, just one sentence in plain language, the way you I'll paste in this prompt. Then, I'll set up the settings, 16 by 9 for the ratio and 1080p for the resolution. For the duration, I'll pick 15 seconds, and
[06:20] high. Now, there are two options at the bottom. You can hit create video, which you can hit preview sheet, which builds a full plan before generating anything. I'll go with preview sheet because this is where Smart Shot actually shows what
[06:34] here's what we get back. The first thing you see is a character reference sheet. For our astronaut, it shows the suit from the front, the side, and the back. the gloves, and the boots. Now, this part is honestly so important because I
[06:49] sheets like this myself every time I wanted to make a high-quality AI video. starts changing from one shot to basically what keeps everything consistent. But, what Smart Shot does is
[07:02] generate this for you automatically. And just the character sheet alone used to take me around 20 to 30 minutes to put together manually. Now, I get it from which is honestly insane. So, let me show you the next thing it created,
[07:15] basically the location where the entire scene takes place. And the reason you want to define this before generating any video is because it keeps the letting it randomly change between shots. Because that's one of the biggest
[07:28] character might look good, but then the But, what I really like about Smart Shot is that it uses GPT image to build a five separate shots showing how the scene plays out, and each one comes with
[07:42] its own camera angle and a short description. For the camera work, scene. So, it might use something like a 35 mm tracking shot for the wide scene, or a 40 mm dolly shot for the close-up. And on top of that, it plots the camera
[07:56] movement directly on the floor plan, so you can see exactly how the camera moves creates the lighting and mood. So, you get the color palette, the type of overall feeling of the scene. And this matters more than it might seem, because
[08:10] beginning to the end. Otherwise, the video feels like a bunch of random shots earlier, these are all the exact elements I used to build myself when I was planning a high-quality AI video. The only difference is that before, it
[08:24] everything. Now, I get the whole thing from one click. But if you don't like something, you can still go in and edit it at any time. So, you're not losing boring manual work. Now, all that's left
[08:37] is to generate the video. So, I'll hit create video, and here's the result.
[08:56] in the reference sheet, and the suit details stay consistent across every shot. The lighting also holds really well with the warm sunlight and the camera moves between the wide shot, the close-up, [music] and the action moment
[09:10] exactly like the storyboard planned. So, this is a 15-second cinematic video generated from one sentence in just a few minutes. And the quality is high straight into a real project. So, if you want to use the new GPT image yourself
[09:23] or create your own AI videos in 5 minutes, I'll leave a link to OpenArt can sign up. Thanks for watching, and I'll see you in the next one.