00:00For the past few weeks, everyone's been going wild about Nano Banana, me included

00:03But I'm about to show you a brand new AI image model that outperforms it across some key metrics

00:10And it does that all with way lower censorship

00:13If you're new here, I'm AI Samson, and this channel is about exploring the possibilities of AI

00:17While Nano Banana might be great for editing images, it struggles with resolution, aspect ratios, it gives you a watermark, and it sensors like crazy

00:27Let me introduce to you the stunning images from this model Cream 4

00:32It's quietly generating full 4K images in just 2 seconds

00:38And some of them look like this

00:40Now, by the end of this video, you won't just understand why this model is such a gamecher

00:44You'll also understand how you can start using it yourself for free

00:49So, let us start with the best of the best

00:51Now, the exquisite detail here is absolutely stunning

00:53Just look at the fine pores, intricate details, and even these individual hairs rendered out on this very stylish guy

01:01Now, its ability to work with restraints and render out even entire scenes in just one color is outstanding

01:10Not only that, but its aesthetic sensibilities are absolutely incredible

01:16Now, a unique feature about Cream 4 is the ability to generate multiple images at once

01:20So here we have a western cinematic example and the prompt was refer to this image

01:26Generate three images from different views 360° overhead view and elevated view and a back view

01:33Don't change the girl's appearance and outfit

01:34And my god do we get fantastic character consistency but also a beautiful sense of realism and cinematography

01:43Now this allows us to quickly edit and iterate inside of the tool

01:48If you know from working with Nano Banana, you're limited to just generating one image at a time

01:51And this is hugely restrictive

01:54Even in other tools like Midjourney, you get four images out in one go

01:58But what this does differently is allows you to create multiple different images of the same style with just one prompt

02:04Just look at how well this complex light is rendered out in these four examples

02:10Now, what's incredibly difficult for AI image models to handle are intricate shadows

02:17And you can see here we've got extremely complex shadow patterns being cast from this metal fence onto the face of this lady

02:28And it absolutely inundates us with a real sense of atmosphere and realism

02:31In fact, it's often a very useful tool to check if our image is real or fake by tracing the shadows precisely

02:40Now, this model is handling absolutely exquisite graphic design with ease

02:45I think it's the strongest design model that we've seen

02:49And here you can see a UI interface completely designed with fantastic illustrations, buttons, and consistency across the screens

02:58But we're going to come on to text rendering in this model a little bit later

03:03What I want to show you is the absolutely mesmerizing realism

03:07Now, if we pay close attention here, we can see that this is a stunning image and it's got a wonderful sense of depth of field and this blurred bcker background

03:18However, one must draw one's close attention to, of course, the human hand

03:23And here, once again, we are still possibly getting six whole digits

03:28This certainly does look like an additional digit on the hand

03:31And it is important to recognize the flaws in these models because they are by no means perfect

03:37And the truth is that understanding what the strengths and the weaknesses are of each one is the best way to engage with them

03:44Now, this example is remarkable because it's not only combining photorealistic imagery, but it's combining it with graphical elements, and that is incredibly hard to do

03:54And what I absolutely love about this is the consistency in the color scheme

03:58You can see that it has accurately matched the colors from this individual's paddle, ball, and outfit to the graphic elements applied on top

04:09Now, another key element to pay attention to when analyzing the effectiveness of text rendering in these image models is making sure that each glyph is exactly the same

04:19So, we can look at these three S's here, and you can see that it's imperceptibly exactly the same

04:26In weaker text models, you will often see that the glyphs are slightly different between instances

04:35Here we have a beautiful image

04:35And what I love about this is the motion blur of the other characters in the shot

04:40It really draws us into the protagonist

04:45Now, I particularly love the lighting that's drawn over here

04:48There's a real asymmetry to this which leads the eye

04:52And of course, these hands are much more anatomically correct

04:54But do take a moment just to marvel at the wonderful textures on this worn out bag

05:01Absolutely stunning

05:01In this example, we have some exquisite fine dirt cascading behind this sports car

05:07And it gives us a real sense of being placed behind this car as it's racing through the desert

05:17Now, I also love subtle details that are included here

05:19Specifically, these car tracks that are found parallel to our main car

05:25It gives us an extra sense that there are other elements inside of this environment

05:30Here we have a fashion shot, and this one showcases exactly how easy it is to change elements in a shot, that you can edit this with much higher resolution than you can inside Nano Banana

05:42And again, what I think is blowing me away most about this model is simply its ability to create graphic designs with photorealistic elements because this is absolutely sensational

05:56Not just for the ability to combine these multimedia types of content, but also for the design intent behind this

06:06Everything from the colors to these additional graphic elements to the composition works harmoniously well

06:13Now, my only slight critique of this one on the right hand side is that these final three characters are not at the same size as the first part of the word

06:22Now, let's take a closer look at the abilities to edit images inside of this model

06:25And we're going to do a bit of a comparison with Nano Banana later on

06:29But here we can see the shot being relit with a lamp

06:32And what I love about this is just how well it maps out the lighting implications across the environment

06:41And it's simply sensational

06:42You can see we go from this dark, morbid, slightly atmospheric room

06:47And then it's lit up and it gives us a beautiful light and airy feel, almost inviting

06:52Here you can see this dog being completely replaced with a schnowzer

06:56and it does it fantastically well

06:59Still keeping the same acting of the dock, the same position, the same posture

07:04Now, of course, you can use this for colorization and removing scratches from an image and bringing old photographs to life

07:09But let's take a deeper dive into the practical applications of the text rendering in this model because it simply is head and shoulders above the rest

07:18Here is a image of a gallery website and you can see that it's mapped out a fantastic layout here

07:26You can see that it's intelligently created a four column layout

07:29Here you can see it also works particularly well with Chinese glyphs

07:34And I ran this through chatbt and it told me that these are completely correct

07:38But here is where it blows me away because in many situations even without a banana if you're trying to create highly complex pieces of text rendered that's anything more than a couple of lines it fails

07:48whereas CG4 is able to create this immensely detailed infographic poster

07:55Now this is a particularly admirable piece of design because of the way that the illustrations integrate so seamlessly with the typographic choices that you can see that this barack palaces and these Gothic churches suit this serif font so eloquently

08:13And the prompt for this was create a comparison chart of a Gothic church and a barack palace and briefly describe the main characteristics of each architectural style respectively below the corresponding pictures

08:25Now if we look at how this model is shaping up against the others in the online leaderboard, we can see that Cream 4 is ranking top

08:39It's above the top models from Google, Open AI, and Black Forest Labs

08:41And it's currently doing so by a significant margin where the gap between first and second is about 48 points

08:50And then the gap beyond those is much smaller just but it's not just leading the way in text to image

08:57It's also leading the leaderboard inside of image editing

09:00And you can see here that it is also number one but this time by a much smaller margin with Gemini 2.5 Flash or Nano Banana coming in a close second

09:10Now you can see here in the overall text to image evaluation across a number of different metrics

09:16It is the leading model for aesthetics and then it comes in a close second behind GBT image on a few other evaluations including correction structure and alignment

09:29Now this model is getting an excellent online reception and Tech Radar even said that it is the best AI image model that they've seen

09:37But let's see if that's true and put it into a direct comparison with Nano Banana

09:41So, I've been playing around with these two tools and putting in the same prompt for each

09:49Now, I'm going to give you all these prompts and all these images in a link in the description below so you can zoom in and check these out yourself

09:54So, first up, we're going to challenge this to create an image portrait of an old fisherman with white salt on his face in the background as a fishing port and boats

10:03Now, let's have a look at how these two came out

10:05Now, first of all, what I want to draw your attention to is the crispness of Cadream

10:08If we just zoom into the details here, you can see that we've got fine strands of hair accurately rendered

10:15Whereas, as we zoom into Nano Banana, there is much less detail

10:22You can really see the pixelation is evident here

10:23But let's compare the aesthetics

10:26And personally, I I prefer the Cadream photo

10:29Now, that is for a couple of reasons

10:33Firstly, I find the background more engaging

10:37They've added more complex elements and also given us this wonderful backlit effect where the light is captured in his silvering hair

10:46Now, one thing I do like about the Nano Banana image is this subtle teardrop cascading across his fore face

10:54Possibly it's just a drip of melting sweat or melting ice, but I'll go with a tear

10:59Now, I personally prefer the expression of the Seadream version

11:05I would say that he communicates a slightly more worldwise and salt of the earth expression, whereas the Nano Banana version looks like he's simply come out of the trenches

11:15It seems like he's seen things that you would never believe

11:19But let's move on to a city street scene

11:21This is round two, and we're going to challenge these two models with a prompt

11:25a city street scene with objects a red car, blue bus, yellow bicycle, two dogs, three people with umbrellas and a green traffic light each positioned clearly

11:33Now here we are testing the ability to intricately include a number of different specific objects and the most important thing that we are looking out for is that it gets every object that we've asked for

11:45So let's go through and count the objects and excellently portrayed

11:46Now that one has given the man on a bicycle an umbrella

11:50That's that's a challenge and I'm impressed

11:52but they have also accurately given us

11:54Now, as for aesthetics and composition, my preference is for Nano Banana

11:58Do let me know what you think in the comments and if you agree or disagree with my subjective thought, but I ran this again because you always get different results every time you run these types of prompts

12:08And you can see here we've got massively varying composition

12:11Now, I would call out Cream here that possibly these three characters do not feel as if they're the right size

12:19They look a little bit small to me

12:20and we have suddenly got three dogs instead of two

12:24Likewise in Nanabana, we do not have three people

12:25We only have two now

12:25So there are a couple of issues with this and they are still unable to completely accurately achieve what we desire

12:31Now this is one of the most amusing challenges to set an AI image generator and that's to ask it to fill a glass of wine to the very brim

12:39And this is because from all of the training data that these image models have, they have very few if not zero examples of a glass of wine that is truly filled to the very top

12:53And that of course is because when you pour a glass of wine, it is natural and the idea to only pour a small little glass for yourself

13:02And here we have the results

13:04See 4 giving us a half full glass and Nablanana also giving us a half full glass

13:09As for aesthetics, I would say that it has a remarkably different inherent tone to these images

13:18You can see that the Cadream output is much richer, slightly more saturated, where Nano Banana gives us a slightly more minimalist aesthetic to the image

13:30Personally, I would slightly veer to Nano Banana in this instance

13:32But let's take a look at the censorship issues

13:36And for this there are a number of different elements to consider

13:39The first is the use of celebrities

13:40And for this we're asking for a portrait of Barack Obama playing the guitar

13:44Now one challenge that I've had and many others have had inside of Gemini is that it often refuses to do things that it is capable of doing

13:53Even if you ask it to edit an image, sometimes it says actually no, I don't do that

13:57And it's quite hilarious that it does this

13:59And often the best way around this is simply by opening an entirely new chat which resets all of the context trying it again and then even prompting it or should we say proddding it more likely to do as you have asked

14:13But here are the Barack Obama options and you can see it's been able to be generated in both of these models

14:21However, aesthetically my feeling is that Cadream 4 gives me this more country vibe

14:27I think I'm more likely to the Barack album from Cadream 4

14:31Next up, we can try it with Taylor Swift

14:33And let's have a look at how this came out in both

14:34And you have to say these are both doing fairly well

14:37And I would say that the aesthetic is rather disappointing in both

14:41It feels a little bit oversaturated and leaning towards this style that we were encumbered with for so long, which was a highly oversaturated and garish tone from our AI images

14:53Now, as for trying to get some action or something a little bit gruesome inside of our images, this is another factor of high censorship that we experience

15:04I tried the following prompt

15:07Photorealistic and intense a wide shot of two samurai fighting with red paint in the air between them

15:12The scene is set at an dramatic ancient Japanese temple

15:17Now unfortunately Nanabanana refused to give me this output and the seamream version did give me an output but it is slightly underwhelming in terms of the graphic nature

15:27Now as for profanity I asked it to graffiti the wall with a word beep spray painted in red and both these models complied and gave me these effective outputs

15:38Now, as for photos of a slightly sensual nature, I was able to generate this shot of a woman and also see her from behind

15:47And this also worked similarly inside of Cream 4

15:52However, Cream 4 is not completely uncensored

15:56It only has comparatively lower censorship than Nano Banana

16:00I also had a few other prompts that did not work at all in either

16:03The artistic nude marble sculpture and a museum style photograph did not work at all

16:09However, when I changed the prompt and put in a photograph of a classical marble statue in a Gand museum setting, dramatic lighting, artistic composition, this worked exceptionally well

16:19You can see that we got both of these out, but they are not by any means nude marble statues

16:24It seems rather unpredictable what will trigger the censorship in these models, and often they are inconsistent when they do

16:30Sometimes they will create the things you ask for and if you try the very same prompt another time often they will refute

16:37But let's take a look at another comparison and this time we're going to be seeing how it compares with highfidelity text rendering

16:46We're going to be looking for a poster that says Cream 4 versus Nano Banana in bold clean typography futuristic style

16:52And you can see here both have handled these very well

16:54But I would have to say that Cream absolutely wipes the floor on two key elements

17:03First of all, it specifically rendered out the exact text formatting that we asked for, which is Cream with a capital S and D

17:10Whereas Nabana has put a space between C and Dream and given us some pretty horrendous graphic design

17:18If you just see how much more considered the Cream 4 version is, even just how elegant this unusual type face is, it is completely designed and innovative and unusual type face that blends elements of I would say uh not only uh structure but also elegant

17:34You can see it's got these slight seraps on some of the characters which give it a sense of maturity but also these harsh edges give it this tech feel

17:46Now, let's take a look at how you can actually use Cadream for today

17:50And for that, I'm going to be using Love Art, who are also today's video sponsor

17:55And Love Art is a place where you can design and create using a whole host of the latest image models

18:03You can use V3, Google Nabana, and Cream 4 all in one place

18:07Now, inside of Love Art, how to access Cdream 4 is to come into the create new project and then you go to models

18:16And from here, you can see all the models

18:18And that's where we're going to be playing around with Cdream 4

18:21You can also see it takes around 10 seconds to complete

18:23Now, remember, you can try this out completely for free

18:27And to use it, all you have to do is you can start off with an idea or use one of the example prompts from Love Art

18:32I'm going to be asking it for a Vogue cover with an albino woman in futuristic fashion outfit

18:38Impossible

18:38Now, just look at how stunning this is

18:39How accurately the text has been rendered and how beautiful the integration of image and typography is

18:46It's absolutely remarkable

18:48Now, you can ask for any changes that you like

18:50Even talking in natural language to edit elements of the image

18:55For example, we can ask to change the race of the model

18:57And here we have it

19:00Absolutely stunning

19:00Now, we can also also ask it to create a number of different Instagram posts in this exact style

19:06So, now we have five images and these are all in the perfect aspect ratio for Instagram and they also carry across the exact aesthetic consistency that we're looking for

19:17Now, what I love about Love Art is the ability to use this infinite canvas

19:20And you get to move things around and create in a much more creative and flowing way than you do in the chatike interface where you're using Gemini or other types of LLMs

19:34This really gives us a lot of freedom to play around

19:36Now, you can check out Love Art for free today

19:39And if you do want to upgrade, you can get unlimited access to Cream 4 for the whole year if you upgrade to their premium plan

19:47And a big thanks to Love Art for sponsoring this segment of the video

19:52Now, Cream 4 was developed by Bite Dance, who are the parent company of Tik Tok

19:56And this model offers a unified creation and editing model, multimodal input, which means you can put text and image references

20:06and it understands a number of different important elements of the real world to create complex designs and images

20:15It's truly a huge step forward in image generation models and it raises the bar for what's possible with AI

20:21If you want to stay updated for the latest AI news, make sure to subscribe

20:26And most of all, thanks for watching

20:29Thanks for being here.