New AI image generator BEATS EVERYTHING

Added: 2026-05-05

[00:00:00]The following images were completely made with AI. Yes, even these posters and this photo, everything was made with AI. Openi just dropped chat GPT images, too. And this is by far the best AI image generator you can use right now.

[00:00:16]It's not even close. So, in this video, I'm going to show you all the insane things that you can do with it. Plus, of course, I'm going to compare it head-to-head with the previous best image model, Google's Nano Banana. And of course, I'm going to go over its specs and how and where you can use this. Let's jump right in. All right, so first of all, Chat GPT Images 2.0 is an image generator and editor. So, you can just enter a prompt to generate an image or you can also upload multiple reference images for it to edit. Let's start off with a difficult prompt already. So, here the prompt is a grid of 100 posters of anime shows or movies.

[00:00:54]Include their names. And here's what I got from GPT image 2. You can see that indeed we have Spirited Away, Your Name, Attack on Titan, Demon Slayer, Fumale Alchemist, Nudo, etc., etc. Most of these do look correct to me. I mean, for most of these, it's able to generate the actual characters and art style from the movie. This is really good. Notice that most of the text is also correct. Now, I'm only able to identify like half of these. Let me know in the comments below if it got any wrong. And then here is the same prompt but using nano banana pro. And actually for this one I also turned on web search so it could actually search the web to give it more guidance on what to generate. But even with web search you can see that the quality of the results aren't as good as GPT image. You can see like most of the faces are messed up for these posters.

[00:01:42]Especially like Goku's face. Nudo also looks messed up. So definitely not as detailed as GPT image 2. And let me just also scroll down slowly so you can see all of these results. This is the weakness of Nano Banana is if you try to get it to generate a grid of a 100 items, it just can't generate that resolution. So most of these faces are messed up and also the text is messed up. So for example, you can see some misspellings over here. This is suddenly 900 and then bell is also repeated twice. Everything is just not as good as the generation from GPT image 2. You can see everything is a lot more clearer and the text and the characters look great.

[00:02:22]All right, here's another incredibly tricky prompt. We have a screenshot of a Windows 11 desktop. It's quite messy with lots of overlapping windows in random locations. One window shows Slack in Chrome, another shows Gmail in Chrome, another window is Excel, and another window shows a PowerPoint presentation on a top secret OpenAI project. So, here's what we got from GPT image 2. Indeed, we have Chrome open with Slack and then we have another Chrome window open with Gmail. And the text within each window actually looks correct. It's not just gibberish.

[00:02:58]Although, one error is that this part over here is kind of messed up. It's not really a straight line. And then we do have Excel open. And again, most of the text within this window are actually correct. It looks like a legit financial spreadsheet. And then here is a PowerPoint presentation. Also, most of the icons down here do look correct. And then if we look at some of these desktop items, they also look mostly correct. So pretty impressive. And then here we have Nano Banana Pro. So we do have a Slack window open. However, you can see a lot more misspellings in this Slack window.

[00:03:32]And then same with Gmail. There's just a ton of misspellings. It's just outputting gibberish. Same with this Excel spreadsheet. This just looks completely gibberish. This doesn't even make sense. There's some misspellings and some deformationations everywhere.

[00:03:45]And then finally, we have PowerPoint over here, but this window doesn't really look like the PowerPoint interface. And you know, strangely, it actually had two windows of Gmail open.

[00:03:55]And then even with the desktop items, like some of these are not spelled correctly. And then down here, it's missing the Excel icon. So, both are not perfect, but GPT image 2 has a lot fewer noticeable errors compared to Nano Banana Pro. Here's another fun example.

[00:04:13]So, we have a screenshot of the YouTube homepage for a tech broker. And here is what we got from GPT image 2. First of all, let's look at the left sidebar.

[00:04:22]Everything does look correct to me, but there's one major error from this generation, which is that my channel is not included in this subscriptions list.

[00:04:31]But anyways, you can see all the text and thumbnails are actually correct. The titles, the channel name, and the number of views all look pretty good. And these thumbnails do look like, you know, standard YouTube thumbnails. I guess one minor flaw is that if you zoom in closely, then the faces of these user icons are kind of messed up. So that's cheapy image 2. Next, if we look at Nano Banana Pro, it also did not include AI search in the subscription list. So that's a fail. And then we have some misspellings over here. We also have some alignment issues over here. So, I'm not sure why it added two videos here in this column. And then this one is kind of lacking a title. And then this title is like shoved all the way down here past this line, which is completely wrong. So, again, just a lot more inconsistencies and errors with this generation compared to GPT image 2. Let me know in the comments what you think.

[00:05:25]All right. Next, we have a screenshot of a Tik Tok live stream featuring a beautiful woman hosting the stream.

[00:05:30]Here's what we got from GPT image 2. You can see all the text is pretty much perfect, including, you know, the mobile icons at the top. Put all these live stream interface icons. Plus, all the comments in the bottom are also correct.

[00:05:43]Like, it's really hard to spot any flaws with this generation. Whereas, if you look at Nano Banana Pro, this kind of looks too fake. And then the design doesn't really look like a standard live stream interface. Plus, I'm not sure why, but the mobile icons are down here, which is also not standard. So, just in terms of overall aesthetics and realism, again, I would have to give the point to GPT image 2. So, as you can see, so far GPT Image 2 is just dominating Nano Banana Pro. Right now, instead of just generating screenshots, let me also show you some actual use cases for this. So you can easily get it to create some branding materials or design ideas for whatever you want. For example, for the prompt, I got it to create a professional brand visual identity system presentation board for an eco-friendly matcha brand called Mist.

[00:06:31]On the left side, it should have the main logo plus logo construction grid with precise geometric guidelines, inspiration mood board, and the color palette selection include hex codes and then also a topography section. And then on the right side, it should contain some brand application mock-ups. It should have business cards, packaging design, shopping bag design, even a mobile app and website landing page design, plus the employee ID card design. And here's what I got from GPT image 2. So, here's the logo at the top.

[00:07:04]Here is the logo construction with geometric guidelines. Now, this part is messed up. These circles and lines are kind of just randomly drawn. And then here's a mood board for inspiration. And then here is a color palette plus the typography. And then over here we have business cards, the packaging design, the shopping bag, and even the mobile app and website design plus an employee ID card. So it's very good at understanding my prompt and actually including all the elements that I specified. And then here we have Nano Banana Pro, which was also surprisingly able to generate everything in the prompt. We do have a logo over here plus geometric guidelines, although these lines are also just random. And then we have a mood board over here plus a color palette with hex codes plus the typography guidelines. And then over here we have business cards, the packaging design, shopping bag, employee ID card, plus the mobile app and website landing page design. So both are not bad, but I would say Nano Banana looks a bit too cartoony. If I were to pick a winner, I would go with GPT image 2. All right. Now, instead of branding materials, you can also just get it to design an entire catalog of products for you. So, let's say you are a clothing brand. Well, you can easily just get it to plan an entire infographic poster with your products. For example, in our case, let's get it to design a minimalist fashion infographic. It's going to be a 7-day weekly outfit guide for women with a soft neutral beige palette, elegant Korean Chinese aesthetic. Each column shows a full body model with coordinated outfit plus small accessory or product thumbnails beside each look, etc., etc. And here's what we got from GPT image 2. So Monday is office chic, Tuesday smart casual, Wednesday is leisure time, Thursday is date look, Friday is sporty active, Saturday is weekend style, and Sunday is home comfort. All of these actually look really good and very nicely designed.

[00:09:06]Its clothing and accessory selections look very professional to me. Plus, it even gives me a nice color palette at the bottom. And then here is Nano Banana Pros generation. It's also not bad. it was able to generate all these different outfits for each day of the week. But I would say in terms of overall aesthetics, I would have to give the point to GPT image 2. Let me know in the comments what you think. Next, here's a fun one here. The prompt is a 5x5 grid pixel art sprite sheet of a princess warrior sprinting then slashing her sword. So, here are the grids for both image generators. Next, I plugged it through a sprite sheet animator and here's what I got. So for GPT image 2, the princess is first kind of sprinting and then slashing her sword. This does look a lot more dynamic and fluid and consistent. Whereas for Nano Banana, the princess is not really sprinting anywhere. She looks kind of static. And then even the sword slash looks kind of strange. So definitely not as good as the generation from GPT image 2. So, if you're looking to generate sprite sheet animations, then it's safe to say GPT image 2 is currently the best model you can use. Now, GPT image 2 can also take in one or multiple reference images. So, what I'm going to do is upload this pretty complex table of AI models with the name, context window, creator, intelligence index, price per million tokens, speed, and latency. A pretty complex table. Plus, there's also a trick row here, which is Metaz Mu Spark, which does not contain these details.

[00:10:44]And then I asked it to turn this table into bar graphs. Make it look amazing.

[00:10:49]So, first of all, here's what I got from GPT image 2. It gave me a nice header plus key takeaways at the top, which is really nice. And then here is the intelligence index, and it does look correct to me. Next, let's move on to pricing. And if we compare it with the original table, again, everything looks correct to me. Same with speed and latency and context window. All the numbers and labels are actually accurate. And then at the bottom, it even groups the models by company. So, a very thorough and accurate and nice design. All right. Next, we have NanoPanana Pro. And it contains a lot of errors and omissions. For example, here it's missing Quen. It also misspelled GLM 5.1 and Miniax M2.7. And then here it added Quinn back, but it's missing Muse Spark. It's also missing Kimmy. And then here, Muse Spark should be blank, but it added a bar here anyways, which is really strange. Also, we don't even have Sonet 4.8. Some of these bars are just really wrong. Again, a lot of misspellings over here. There's just way too many errors for this generation.

[00:11:55]Whereas for GPT image 2, it was actually able to get all the numbers and labels correct. So, you know, with GPT image 2, maybe you don't even need Excel or any other charting software anymore. You can just take a screenshot of your data table or even just enter the raw data into your prompt and then get GPT image 2 to generate whatever visualization you want. This is a super flexible tool and it's great for creating infographics and visualizations. Now, actually, I already had access to this last night, so I made this post yesterday. But unfortunately, I only got like 400 likes and 61K impressions, which looks pretty awful.

[00:12:35]So, let's fake the numbers a bit. I'm going to take a screenshot of this and then plug it into the image generator.

[00:12:41]And then for the prompt, I wrote, make this post have over a million views, likes, and comments. Also add some comments below it. So, here's what I got from GPT image 2. It was able to indeed increase my comments and resharers likes etc. Although one error is the number of views is lower than the number of likes which is not correct. And then it even had the CEO of Perplexity I think comment on this and even Mr. Beast left a comment. And then here is Nanogano pros generation. Here we start to see some huge errors. So for the comments and reshares the format should not be like this. So this is completely wrong.

[00:13:19]And then the text and the format of the comments are also not really correct. If you compare the comment section from GPT image 2, you can see that this just looks a lot better. So if you want to fake some numbers or if you want to generate some fake screenshots of whatever, then GPT image 2 is by far the best model you can use right now. All right, here's another fun use case for this. Let's say you see some font which you like. Well, you can just ask GPT image to generate the entire topography design of this. So, here I wrote include upper and lowercase and numbers. And here's what I got from GPT image 2. The typography does align with the input reference image very well. And then here's what we got from NanoPanana Pro.

[00:14:04]I would say this doesn't look as accurate as GPT image 2. The letters should be a bit wider and also some letters like the lowercase W is not correct as you can see here and the L is also not really correct. So in terms of retaining the style and consistency again GPZ image 2 is noticeably the better model. Now this is also great for creating diagrams. So let's just input this photo of the iPhone 17. I think for the prompt I wrote explode this device down into its separate components and label each black background. Here's what I got from GPT image 2. It was able to explode all the components. So here we have a ceramic shield front glass plus adhesive layer and then this retina display etc etc. Plus we have a thermal management system and a logic board with a A18 chip plus the battery. Although the label of the battery is not correct.

[00:15:02]It should be pointing here. So, not perfect, but overall not too bad. And then here's the generation from Nano Banana Pro. It was also able to identify that this is an iPhone. This looks a bit cartoony. And then the front camera system plus these rear cameras are also not correct. I'm not sure why it added five lenses here even though it knew that there's only like two to three lenses at the back. So, this part is also not correct. And then this USB lightning port should be pointing this direction because the top is over here.

[00:15:35]So there's a lot more inconsistencies with Nano Banana's generation compared to GPT Image 2. None of them are perfect, but if I had to pick a winner again, I would have to pick GPT Image 2.

[00:15:47]So far GPT is just dominating every round. If you want to level up your content and start pumping out high converting ads at scale, definitely check out Higsfield, the sponsor of this video. They just launched Pigsfield Marketing Studio, which is like a full AI ad production pipeline compressed into a single workflow. No script, no agency, no production team, just one prompt and you've got an entire campaign ready to go. Here's how it works. You can paste in a product link or upload an image and the system will automatically generate ads across nine different formats. Everything from UGC style videos and tutorials to cinematic TV spots and even virtual tryons. Instead of testing one or two creatives, you're instantly testing multiple angles, hooks, and formats all at once. What's really impressive is their seed dance 2.0 engine. You can upload your own face or generate an avatar, and it stays perfectly consistent across every single video. Same character, same look, no weird drift. That means you can basically create your own AI brand ambassador and scale content like crazy.

[00:16:54]And the use cases are huge. If you're running e-commerce or drop shipping, you can paste in dozens of product links and generate hundreds of ad creatives in minutes. If you're doing Tik Tok shop or Amazon FBA, you finally get enough volume to properly AB test. Even agencies can use this to deliver full campaigns in under a day at a fraction of the usual cost. Whether you're launching a product, running ads, or building a brand, Higsfield Marketing Studio is easily one of the most powerful tools out there right now. Try it out using the link in the description below. Now, this is also great for creating comics. Here's just one example where I input these two photos and then I asked it to create a black and white manga page. These two characters are having an epic fight. And here's what I got from GPT image 2. Naruto says, "Let's settle this." I guess it was able to identify that this is Nudo. And it's even able to make him generate this move. So that's the generation from GPT image 2. Notice that nowhere in the prompt did I specify that this is Nudo or this is Gojo. And then here is Nano Banana Pro's generation. It was also able to identify that my input image is Nudo and then get him to do this move.

[00:18:06]So it's a close call. I would say both image models were able to generate a decent looking manga page of these two characters having a fight. So now it's really easy for anyone to generate comics with whatever you want. You just need to input reference images of your characters and locations and prompt it on what exactly you want for each page.

[00:18:28]Heck, you can even just get an AI agent to plan out each page and then you just plug the prompts into one of these models. All right, here's another great use case for this, which is getting it to redesign certain things. I can plug a screenshot of my website onto here and then get it to redesign this landing page and make it look better. So, here's what GPT image 2 gave me. It completely redesigned the header plus the tags over here. And this does look like a very nice design. And then here is the generation from Nano Banana Pro. It's not too much of a redesign. It added some highlights and shadows and added some color gradients, but I would say GPT image 2 has the better design. Let me know in the comments what you think.

[00:19:09]Here's another great use case for this.

[00:19:12]You can just plug in a photo of any product and get it to make a storyboard for an ad for that product. In my case, these are lightweight noiseancelling earbuds. Be creative and make it very appealing. Below each scene, provide a description. So, here's what we got from GPT image 2. It starts with this dude walking in a crowd overwhelmed by the noise of the world around you. And then we would zoom into his face wearing these earbuds. And then it would show this nice visualization of the earbuds canceling all this ambient noise. And then we move on to the next scene and so on and so forth. And then here's what we got from Nano Banana Pro. This is way simpler and a lot worse than the storyboard from GPT image 2. So again, in this example, I would have to give the point to GPT. All right, next. Here is a really tricky prompt that not a lot of image generators can get correct. A bustling street scene in Hong Kong with signs in Chinese and English. This prompt tends to trip up a lot of image models. They were not able to generate signs with legit text. There's a lot of misspellings or they just generated gig.

[00:20:19]But here is what we got from GPT image 2. Most of these signs and logos actually look correct. This street sign is also correct, although there are still some misspellings, especially for some Chinese characters, like this one over here. And then there's some gibberish over here. And most of this text is also wrong. So, it's not perfect. If you look closely, there are still some errors with this generation, but it's already really impressive how it knows all of these different signs and logos from different companies. It's also impressive how it's able to identify what a Hong Kong tram looks like as well as a Hong Kong bus. And then here is the generation from Nano Banana Pro. This is also very good. It's able to generate signs with legible text. However, some of the signs near the back are still gibberish. Here you can see some misspellings for currency exchange and then also some misspellings over here. So, none of them could ace this prompt, but honestly, both generations are very good. And from far away, this looks exactly like a street scene in Hong Kong. In this case, I would have to say it's a tie. Now, since GTA 6 ain't never gonna come, I also asked it to generate a gameplay of GTA 6. So, here's what I got from GPT Image 2. You can see all the text is accurate.

[00:21:37]The map, the cars, everything looks very similar to an actual GTA game. And then here's the generation from Nano Banana Pro. also not too bad, but there's a lot more errors with this map interface over here in terms of having the least amount of errors and being the most accurate and just looking better. I would have to give the points to GPT image 2. All right, next. Let's see if it can do your homework. So, here we have a biology worksheet where you need to label all these organels in an animal cell. For the prompt, I wrote label these organels. Use a messy students handwriting. All right, here's what I got from GPT image 2. It actually was not able to get a lot of these correct.

[00:22:16]So mitochondrion is correct. Cell membrane is also correct. But it got some of these wrong like nucleus and nucleololis. Rough ER is also wrong.

[00:22:25]This ribosome is also wrong. It even left some of these blank. So in terms of its biology understanding, it's not too good. Next here is what we got from Nanopanana Pro. I would say it got even more wrong. It labeled this mitochondria as cell membrane and it labeled the cell membrane as nucleus. This is just an absolute fail. So honestly, both are not great. It doesn't seem like they have a built-in knowledge of, you know, cell organels and what they look like. All right, here's an even tougher prompt. A 3x3 grid of endemic frog species of Borneo. So these are frogs that are only found in Borneo and nowhere else in the world. Below each photo show their common name, scientific name, and a brief description. This is just a typo.

[00:23:09]All right, so here's what we got from GPT image 2. And I'll save you the trouble of googling each of these, but basically it got all nine of these wrong. These images don't look like the actual frog. Plus, some of these are not endemic to Bonio. For example, the Bornean horned frog is actually found elsewhere in the world as well. It's just generating some random looking frogs, but the ones in real life don't look anything like these photos. So, that's a complete fail. Next, we have Nano Banana Pro. It actually generated some better looking photos. So, here we have the Wallace's flying frog, which does look like this. Here we have the Bourneian hornfrog, which also looks like this. However, both of these are not actually endemic to Borneo. They're found elsewhere as well. So, both of these are still wrong. And then again, the rest of these images don't look like the actual frog in real life. So, it's a complete fail for both image models.

[00:24:04]None of them were even able to get one cell completely correct. Maybe we'll have to wait for Gigab Banana or GPT image 3. All right. Next, let's test its understanding of geography. So, we have a highly detailed world map showing Earth's topography with elevation, continents, countries, major mountain ranges, and oceans labeled clearly. At the bottom, list the top five largest countries include area. Largest mountain ranges include elevation, and most populated cities, including population.

[00:24:35]So, here's what we got from GPD image 2.

[00:24:38]The map does show elevation. It's able to label all the major oceans. However, it failed to label some continents. For example, it seems like North America was never labeled here. And then there are still some misspellings for some countries like over here. And then it seems like some countries are not labeled. Next, here are the top largest countries including the area. And then the top largest mountain ranges by elevation. Top five most populous cities. Really, Tokyo is the most popular city. Well, at least according to Google search, it seems like Jakarta, Indonesia should now be the world's most populous city. And Tokyo has fallen to third place. So, this seems wrong. In fact, Jakarta isn't even on this list.

[00:25:22]And then here is Nano Banana Pro. You can see a lot more misspellings, especially for the country names. It failed to even label the mountain ranges correctly. There's a lot of gibberish in the country names over here. It couldn't even label the Himalayas correctly. Just a lot more errors in terms of text. So, from those errors alone, it's safe to say that GPT image 2, even though it's not perfect, is the clear winner in this round. All right, here's an even trickier example. I want to generate a map of Hong Kong in correct geographical proportion in dark mode with the MTR subway lines highlighted and overlaid on top. Label all stations in both Chinese and English. Here's what we got from GPT image 2. It was able to label all stations very well, which is already incredibly impressive. There are some errors with where the lines are drawn.

[00:26:13]For example, these stations are completely wrong. And then there's also some errors over here. But the fact that it's able to generate the names of all these stations, plus roughly the correct location, plus the correct line color, plus it's even able to give me the names of each line, is already incredibly impressive, especially if you compare this to Nano Banana Pro. Everything is wrong. It's missing like 80% of the stations. This looks absolutely horrible. So, without a doubt, in this case, the clear winner is GPT Image 2.

[00:26:44]Here's a really interesting test on its spatial understanding. Here I have a floor plan of this room and I want it to generate a realistic photo of this room taken from the main door. Now the main door is over here. So ideally it should generate a photo taken from this perspective. Now unfortunately for GPT image 2, it completely failed to do this. It generated an image facing the bed which is like somewhere over here which is not correct. Whereas for Nano Banana Pro, it was actually able to get this correct. Still some flaws with this generation. For example, the bathtub should not be facing this wall. It should be facing the left wall like this. It's not perfect, but this is still better than GPT image 2, which couldn't even understand where the door is from this image. All right, next up, here's an even trickier prompt that no imit model is able to get correct so far. So, we have a chess board midame where black is in checkmate in two moves. Show the next moves to reach checkmate. So, here's what we got from GPT image 2. And this is already very wrong. Here it says that the next move the queen should move over here, but the queen actually can't move over here. It can only move to this cell, in which case he would be eaten by this guy. So, this is already wrong. And then here's what we got from NanoPanana Pro. It doesn't show me the next two moves. And this board configuration doesn't even look correct. So, I think both models still could not generate chess games very well. All right, here's a really tricky prompt that no image model has gotten correct so far. 11:15 on the clock and a wine glass filled to the top. Here's what we got from GPT image 2. It is showing 11:15 plus a wine glass filled to the top. Finally, we have an image model that can get this prompt correct. As you can see, Nano Banana Pro still fails to generate 11:15 plus a wine glass filled to the top. So, very impressive performance from GPT image 2.

[00:28:47]Finally, here is a Where's Waldo test?

[00:28:49]So, I got it to generate a Where's Waldo image. Here's what we got from GPT image 2. If we zoom in, notice that the details of the people are pretty bad.

[00:28:59]They all just look like squiggles. So this is not very well defined. And then here is Nano Banana Pros generation. You can see that it has added way too many Waldos in this image. So that's a fail.

[00:29:11]So currently both models fail at generating a good-looking Where's Waldto image. So that sums up some of my tests.

[00:29:18]For over like 90% of the instances, GBT image was clearly better than Nano Banana. Here are some of their showcase generations. You can also easily generate magazines or newspapers about any topic you want. All of this text is actually accurate. Or here is a messy handwritten essay. This looks so damn realistic. I mean, it's really hard to tell that this was made with AI. And then, like I showed you before, this is really good at generating comics while keeping all of the text consistent within all the panels. It's also great with typography. So, here's another example showing characters of different languages mixed together in this design.

[00:29:58]And it's also incredible at creating photorealistic images. Notice the face is not perfect. Her hair is everywhere.

[00:30:04]It has some nice natural imperfections.

[00:30:07]This is really good. Now, to be fair, even the top open- source models out there like Flex Klein or Zimage can already generate really photorealistic images. So, you don't have to use GPT image for this use case. Here's another challenging prompt where we have a photograph of a professor giving a presentation about GPT image 2, which goes on recursively forever. And again, this looks insanely realistic. The awesome thing about GPT image is it can generate images in non-standard aspect ratios as well. For example, you can get it to generate a 3:1 ultra wide aspect ratio of a time-lapse of a dude doing a slam dunk. Or here's an iPhone panorama shot in a busy Asian city. Here's another example where we can get it to generate an ultra-long image. So, here is the image. Pretty cool. And like I showed you in my tests, this is really good at creating infographics and data visualizations like this. It's also great for designing marketing materials, posters, ads, storyboards, etc. I mean, forget about Canva. Forget about hiring any design agency. GPT image can just output these beautiful professional designs in seconds. All right, so those are some of their examples. Next, let's go over how and where to use it. So here they say that chat GPT images 2.0 is already available right now for all chat GPT users. So on chat GPT, even if you're on the free plan, you can simply click on create an image and then enter your prompt here to create an image. Now you can also specify the aspect ratio by including it in your prompt. For example, we can type like 16 to 9 or in my case, let's do 1 one like this. And let's press run. And here's my result.

[00:31:51]Now, if you're on the free plan, I think you get like roughly 3 to five tries per day. And then it resets the next day.

[00:31:58]And then if you go for a paid plan, then you get higher limits. And alternatively, GPT image is also available on other thirdparty platforms as well as through their API. Finally, let's go over the specs and performance of this. First of all, note that this can generate images of up to 2K resolution through their API or through a third-party provider. If you just use it natively in the chat GPT interface, especially if you're on the free plan, then you can only generate 1K resolution images. This model also has much stronger multilingual support with significant improvements in non-Latin text generation, particularly in these languages and beyond. So, this is great if you want to translate certain materials or posters into another language. It's also incredibly good at photorealistic images. So, it's able to include like natural imperfections as well as tiny flaws that you often see in real photos. And then in terms of aspect ratios, this can go as wide or as tall as 3:1 or 1:3. So, you can create panoramic images or banners. Note that it has a knowledge cutoff of December 2025. This is you know the knowledge that it has built into the model but in addition you can also couple this with a thinking model in chat GPT. So this is kind of like an agent which can use different tools like web search to fetch the latest information. This can help you fetch the latest information that's beyond just December of 2025. And the nice thing about using this natively in chat GPT with the agent mode turned on is that you can create multiple images at once. For example, a sequence of manga pages, a set of redesigns or like content in different aspect ratios or different languages or different designs. Now, if you look at this independent leaderboard called Arena, where people can blind test different AI models side by side, you can see that GPT image 2 by far dominates both text to image and image edit. And the lead is crazy. Like, this leads by almost 300 points, whereas the other models are only around 1,200. Same with this one.

[00:34:06]It leads by over 100 points. In fact, that table doesn't really do it justice.

[00:34:10]So, here's a better visualization. And as you can see, even just the medium version scores way higher than the nano banana models. It dominates everything else in all these categories. From 3D imaging and modeling to art, cartoon, anime, fantasy, photo realism, portraits, product and branding, design, text rendering, multi-image edit. This is just an absolute beast. Anyways, that sums up my review. This is by far the best image model you can use right now.

[00:34:39]And the best thing is you can try it for free right now in chat GPT. Let me know what you think of this and what other crazy things were you able to get it to generate. As always, I will be on the lookout for the top AI news and tools to share with you. So if you enjoyed this video, remember to like, share, subscribe, and stay tuned for more content. Also, there's just so much happening in the world of AI every week.

[00:35:03]I can't possibly cover everything on my YouTube channel. So, to really stay uptodate with all that's going on in AI, be sure to subscribe to my free weekly newsletter. The link to that will be in the description below. Thanks for watching and I'll see you in the next one.