Other AI Image websites are Missing this ONE Feature!

Hello everyone. I welcome you back to the Matvid Pro AI YouTube channel. I know I haven't posted in a while and I do apologize for that. I'll be totally dead honest with you. I live in the us it's election season and I have totally been distracted by all of this insane election stuff. So my apologies for not uploading. Also, it looks like the AI news did kind of take a little bit of a break for, you know, the beginning half of this week. And anyways, I want to dive right back into where we left off. In my last video I talked about something called Recraft. It's an AI image generator. Now, as some of you might remember, when I initially spoke about this generator, I mentioned that I really wasn't quite satisfied with its prompt coherency, especially in comparison to Ideogram, AI or even Dall E3, and I still think that's the case. But even having said that, I've spent some time now with Recraft, specifically the website and some of the features this model offers. And I have to say there is actually quite a lot of value beyond just prompt coherence here. So let's jump right into it. When you first sign up to Recraft, this is what you will be greeted with. Obviously we have a canvas style image generator going on. And at first when I saw of this, I was a little bit overwhelmed. I'm like, do they really want to introduce brand new users like this? But I gotta say, I do like the interface. I'm not gonna lie. After using it for maybe an hour, two hours, it's smooth, it's fast, it's clean. I think it has some benefits even over Ideogram. They show off some of the pretty basic capabilities that you have here in Recraft though. And by the way, yes, you can do a certain amount of free generations with this. You can see you get 50 free credits daily and then they have different plans going up from there. First of all, generate photorealistic images. Pretty much every image generator worth its salt these days can do that. But they also have an background remover. And this might seem like a simple Photoshop style tool, but Ideogram doesn't have this. And it's something that is sorely missed. In fact, this is the one tool that I'm missing from Ideogram that prevents me from fully switching from Photoshop to Ideogram for thumbnail creation. You can also edit your own images. You can just drag and drop them in, which is good to see. Ideogram also has that They've got color palettes just like Ideogram, which is cool. They definitely must have followed Ideogram's lead, I think, in introducing this feature. But honestly, the color palettes can be a lot of fun, especially if you're trying to do something very specific. Not a feature I use a whole lot, but it's pretty cool. They've also got a built in upscale or most image generators do have a built in upscaler these days. And of course they do have inpainting. But what's interesting here is I think their best feature isn't even presented in this initial doc that they sort of start you off with. The most impressive feature I have seen from Recraft so far is this it. If you go up to the top left hand corner here and you click Image, you'll then be brought to their regular image generation interface. Right? You got all your settings here, as you typically expect. If you click the model though, to change what model you're using, you'll notice that they have a little plus button here. And you can actually set and make your own custom models based off of image uploads. And this is absolutely huge. Let's start off with a quick example. Let's start off with this image right here. This is an old Dolly 2 generation, if you can believe that, that I am very fond of. It's one of my first generations. And you can see it's. It's like a cool pop Art Lemon character. But it's a very unique style. I think we can agree on that. And by the way, for the custom styles, we can do up to five images to reference from. And it will automatically categorize based off of image recognition, I presume, what category your new style should be in. And then of course, you can do some tests. I'm not gonna bother doing any tests. Cause I already know it works. Well, let me just show you the power of this style transfer. Here is the original Dall E2 generation over on the left. And we will do a new generation with the style dancing watermelon character wearing a VR headset. So this is, you know, another theoretical character that's going to be in the same exact style. Maybe they're grooving and dancing together, who knows? And without training any new models of my own, Bam. It gives you a character that is very much not only in line with the original colors, but also the original art style. And the these characters, I think they very well could be from the same universe. Let's say. Now there's going to be Some specific differences with the models, of course, because this is originally a Dall E2 generation. So, you know, we've got some weird squiggles down here. It's a little bit fuzzy. Obviously, Image generation has improved quite a bit since Dall E2, so the lines are smoother, but it still kept a lot of that same artistic style. And honestly, guys, this is something that a lot of image generators struggle with. Flux can do this very well. Yes, of. But you need to train flux, and it takes time. It takes like 30 minutes at least to train a flux model on images, and this does it instantly, and it works pretty darn good. So let's make a whole cast of dancing characters in this same exact style. I want to do a dancing blueberry wearing a VR headset to kind of match everything. Now, the reason this, I think, in my mind, is a little bit more difficult of a prompt in comparison to the watermelon is because it's a blueberry and there really isn't much blue in that original reference image. So I will be very interested to see how it actually handles this. And there we go. We have two results here. I think I like this one a lot better. It did a great job. It stuck to the color palette. This could be interpreted as a cherry, maybe not so much as a blueberry, but it stuck by the stylistic guns, so to speak. And that's what you want from something like this. Let's come up with a real world scenario to apply to this. Let's say I'm starting a new clothing brand, right? And these are gonna be the characters that represent it. We want them to all match stylistically. They need to be a part of one cohesive unit and look like they belong together. And dang, recraft just nails it. I've never seen another image generation model that doesn't require training that can just reproduce this result so easily. You know what, let's actually head over to Ideogram AI and see if we can use their features to achieve something similar. I don't think we'll get very far, but let's. It's worth a shot. So let's go ahead and upload this same exact character. Right now, the only options we have here are upscale, describe, or remix. Obviously, remix is the feature we're gonna go with. We'll keep the image weight pretty low, and we'll do the same exact prompt. So in this case, you know, we'll take the dancing watermelon. Let's see how it does. And as you can see, it's just not the same. It doesn't have that instant style transfer capture capability. I mean, not even close. If I up the image weight to 80, do we get any closer? I mean, not really. It's just sort of trying to turn lemon into a watermelon sort of. And these are cool images, don't get me wrong. And obviously you know this isn't exactly what the image weight feature here in Ideogram is intended for. I'm just trying to point out that this is not very easily replicatable. Alright, so this time around I've uploaded a max of five images here and these are real film photos that I took quite some years ago, but it was shot on real film. Now the goal here is to replicate this custom style. So we'll say like Kodak Film or something like that. Obviously automatically put into the category of photorealism. Let's add this new style and oh, it seems like we're dealing with some issues. Task failed. I'll be real guys. In all of my testing with this platform I've noticed a lot of hiccups like this where it just seems to have random server bugs or errors that prevent things from being saved. I prepared this style earlier on a separate account, so let me go switch over. Alright folks, we're back. As you can see we have the old Kodak film. It's those same images I already showed you uploaded as its own style. And there are plenty of examples here that I had already generated to show off. This is the first image I made, obviously a photo of a dog, but look at the realism and it really does capture the film aspect. Just as an example, I'm going to pull up some of the original images. Obviously on the left hand side here we have the AI generated image and then here is one of the real ones you can see there's a lot of that fine grain that you'll see in film photography that is also present in the AI generated image. It's sharp and the depth of field seems pretty similar honestly. Although of course you could prompt for different depth of fields. And if you don't know what depth of field is, it means basically how blurry is the background in the simplest of terms. I think one difference that's pretty notable is a lot of the real film photos are a little bit less sharp than the ones that are AI generated. Even though this I would still consider is very filmic blurry esque, especially for AI generated image. But just to compare a landscape photo with a landscape photo, obviously the real one being here on the Right. This is probably one of the most impressive examples. I mean, if you, you zoom right in here, it's honestly difficult to tell, you know, which one is the AI generated image. This is a photo of the aurora borealis shore, so that might give you some inclinations. And this was shot in my backyard, but that grain, that slight softness is definitely there. If I switch back over to that last previous image, we can see there's some stretching going on in the corner of the image. And it was able to even capture that, I think, a little bit. You can see it on the mountain, maybe, and towards the ice. And that is typically a production of lens artifacts in photography. And that's something very small that the model was able to pick up on and replicate, which impressed me. And you can see this looks like a long exposure too, which is actually a very realistic scenario for taking a photo of the Aurora borealis on film. The stars are slightly slanted, showing that the exposure was open for a long time. And for those of you who don't know much about photography or film, it's just impressive that it's able to recognize and replicate something like that just based off of images alone. It might have also picked up on the fact that a lot of photos of the Aurora borealis have these stretched looking stars, because most aurora borealis photos are long exposures. But regardless, I think it's very convincing. And I think it did a great job at transferring this, you know, filmic style in a realistic and impressive way. Here's where things start to get a little bit more fishy, though. Obviously I think it still replicated the style quite well. This was an old orange tabby cat sneaking up to pound on a black kitten in a 90s house. 90s house looks good. We have both of our subjects here. I think, though the cats look a little wonky. And this is where we sort of start to see some of the limitations that come with recraft. His ear is way too big here. Where's his other ear? This cat's eyes are looking a little bit funky from afar. It doesn't look too bad, but if you zoom in, little bit disappointed in that regard. It really seems to depend on the type of content you're trying to generate. This is just whale watch. Simply like really prompt. And I think it nailed this. It looks like an old film photo, especially the way that the colors are tinted and we can see the aberration on the white parts of the image too. That's like another dead giveaway of like an older style lens on an old film camera. You know, obviously we're going to have the grain here as well. And also with the waves, you notice a slower shutter speed. Again, I'm probably mentioning a lot of like camera terms that you guys might not know, but just know that essentially I'm pointing out little details that it was able to pick up on from the photos I submitted and then subsequently replicate. That's what's impressive. Even the vignetting on the sides, like as it sort of starts to get darker around the frame, like man AI is so impressive. All right, so obviously the more realistic direction, you know, a whale watch the Aurora Borealis. These are easier prompts. Once you start asking for more specific things, that's when this model really starts to experience issues. I think that it still did this style quite well. It captured the style quite well in these images, but the prompt adhesion is not very good at all. The alien flying saucer landing in the backyard by the swing set, Aliens stepping out of the ship, flashlights shining at them, and we have major subject confusion going on, which is a little bit disappointing. I think if you train a flux model in these images, you would get much better prompt adhesion. So while this is great for quick and dirty stuff, like realistic stuff, something more complicated like this, if you want it in that style, you're better off training a flux model. The aliens are not coming out of a flying saucer. They're kind of on the swing sets. They don't look very real either. They've got some like weird AI ness about them, which I don't quite like. Definitely an issue. Definitely an issue. I've got another example. Again, this is a very similar prompt. I was doing some different prompts to see if I could get it correct. But again, it's just trying to put aliens near a swing set. We have no flying saucer whatsoever. We've got random bars in the background as well. But again, it has like a really nice film style to the image as a whole. You can see everything's very soft and sort of mushing in the background here. The colors and the way that the light again on the highlights is reflecting off there. It's, it's, it's nice. Even some of this distortion in the bokeh here towards the bottom, which again is characteristic of those old film lenses. I think we did have another example for this one as well. And see, like we didn't even get aliens. This is just a scary looking photo of like these metall humanoid figures. You know, it looks like, you know, bodies that have been encased in metal, and they just put. Put it near a swing set. That's scary. This is pretty much the closest I was able to get. Unfortunately. I was trying to get the background to be at nighttime, which it did not want to do. But, you know, we got a flying saucer that's connected, Looks like it's a part of the swing set. And then we've got, you know, an alien and a person looking like they're about to get abducted by a saucer. It's just not great with prompt coherence. What's really nice is that even though, you know, we used some old film photos to, you know, instantly train this thing to produce results in that style, it still has the ability to do some weird things. And it does have good text coherency, I'll give it that. Obviously, these lemons turned into balloons, which is a little strange, but this. This one came out pretty nice. This is what I was expecting, more or less. I also tried this film style on some very specific famous characters. We've got a party, and it's Batman and Joker together again. It really did a great job with the overall style. It looks like a real film photo, but is that Batman? I mean, that's definitely Batman for sure. Especially with the things on the arm. That's nice. His belts looking a little bit. Not all the way there, though. We don't have the Batman symbol anywhere. And Joker, I think this is, like, classic 90s joker for sure, but he's. His arm there is a little, you know, lacking in detail, I would say, overall. So that's something that they might want to look into in their future models, is getting the level of detail that we see from Ideogram, even midjourney. It's interesting. It's interesting for sure. So let's go ahead and dive deeper into this. Let's actually try me as a style upload a photo of myself. If we say image of a man, is it going to just take me and reproduce an image of me, or is it going to be in the style of, you know, the camera I use for my videos? Oh, okay. So definitely did not make an image of me. We can say that for certain. It's definitely leaning more towards just grabbing the overall artistic aspects and style, which is consistent with what we saw before. So here's another big difference between this and training your own model for flux or something like that, or training a dream booth model. It really is doing style transfer at its core. So I popped in just like a screenshot of my Discord server and I did chat log explaining why Shrek is amazing. And you know, of course it's not gonna be able to do the really small, finer details here, but it definitely put it in like a very Discord esque image. And it's just Shrek is amazing. Everyone's talking about how great Shrek is. It's nice because you could just do infinite numbers of these instant style transfers for really anything you want. I'm uploading this Breeze Copilot logo. Now we'll do Fancy AI logo. That's what this style is gonna be called. Okay, you can see that the spelling was a little bit difficult here, but it mainly focused on just capturing the font sty style and the colors. Alright, so that's the style transfer stuff. I think it's probably the coolest thing that Recraft has to offer. But I do want to get into a few more of their features because there's definitely some interesting stuff. This is a first for me. They have a vector art style and by the way, yes, this can apply to your own uploaded images. But if we do vector art of, let's say a very cool and chill lemon relaxing on the beach, this image we just generated is not just your normal png that's like a thousand by a thousand pixels. If we zoom in, you're going to notice something. This is actually vector art, which means it has quote unquote infinite detail. It's a different file format than your typical JPEG or PNG that you see online. But this is typically done in programs like, you know, Adobe Illustrator for logos and stuff, because you can zoom in and blow it up as much as you want and get infinite details. It will never be pixelated because it's not actually made of pixels. But I've never seen an AI image generator that can generate in this file format and actually generate real true vector art. When I first saw the vector art style on here, I thought that it was just making images that looked like vectors. Nope, it's making real true vector art and that is super cool. That's actually really good for designers because you can start by generating off of a base and then improve upon it in like I said, something like Adobe Illustrator. Right. And I could see people paying plans just to get access to this. But I do believe they also have an API. So you could like build an app that is potentially just a plugin for Adobe Illustrator that generates right inside of there, which would be even better. So that's something really cool that they did with this model and Vectors can be a lot more flexible. Like for example, this is an overall perfect image, but you could blow it up and put it anywhere you want and it would be tack sharp. They've also got presets like product photo. So I think this is just supposed to be like designing mock ups, but I think Ideogram does that quite well. Image of an AI focused T shirt featuring futuristic lemon characters. Something really basic. And I mean, sure, you know, this is definitely a decent enough mock up. It's in that same style, but you can get better detail, better coherency with Ideogram. I'll try that same prompt in Ideogram just to show you. I think Ideogram here is just way more targeted. I think it's more consistent and coherent. Like that looks like a real picture of a shirt minus the tag. Maybe up here. Same thing for this one. This is just creepy. I'd wear it though. I'd rep that anyways. Yeah, this is a little bit more basic, a little bit more stable Diffusion XL feeling in comparison. They've also got some other features like you can do a whole image set at once. So, you know, this is our first image. Second, you know, and it goes down. Let's make a quick little collage with six generations. You know, that's pretty cool. And it does keep them all in a consistent style so they look like they belong together in a collage. And I like that this isn't necessarily new AI technology making something like this possible. Although I think the fact that they're all very consistent in their art style is new AI technology that they developed with the model. But just like adding extra features like this in, I think it does make a difference. I think it does. I think this really is, in comparison to Ideogram, more of like a Photoshop style competitor because they're adding tools like the ability to just put text on the screen, removing background directly, generating vector art. There's definitely some stuff going on here that you just don't see from other AI platforms with this canvas esque canvas style generation. You know what, let's actually try some quick background removal on this guy. We just click remove background. Don't know what model is doing this under the hood, but hey, there you go. That's pretty good. And I could just replace this pretty easily. They actually have a replace background option. We'll say just like deep blue. Oh, it did not want to give me a deep blue background for some reason. It's probably because I removed the background and then tried to replace it. Let's Try it on Shrek over here. Blue background. Yeah, so there you go. It actually works pretty well. I'll tell you though, one thing I really am missing is the ability to do out painting. I understand they do have inpainting, right? So a modify area where, you know, I can select Shrek's head here and turn it into a pumpkin. But while I can do inpainting and I don't think the inpainting is all that bad, I can't do out painting. And out painting is. Well, it's definitely something I do quite a lot in. Not only Photoshop but also Ideogram AI. So you know, the inpainting, it ain't bad. It ain't bad at all. But Ideogram's also got inpainting plus the outpainting, so that's also something to consider. Apparently you can also just convert any of your AI generated images into vector art, which is pretty cool. So real quick, let's try the Vectorize. Okay. I mean, it definitely worked. It's far from what I would call perfect, but I assume that's how they're doing the vector AI image generation in the first place. Looks like you just convert whatever you want, but maybe they have a special fine tuned model to make better vector images. And before we go, guys, the last thing I want to talk about is the ability to see what the community is doing. A lot of different places have this now, you know, midjourney. Ideogram. All the AI image generation websites try to have some form of image sharing because it's great. I can't help but thinking though that their community section here is a little bit lacking. You don't seem to have the ability to see what other people use for their styles because it seems like you can do public styles like the ones that I showed you with film or whatever it might be. Apparently those are defaulted to public. But does that mean you can share them? Because I think sharing styles would be way cooler. Maybe you have to upgrade to see that feature. Let me know if you guys have the deets on that because I would be interested. But the basics are there, so let's talk conclusions. I think in my original video I was maybe giving Recraft a little bit of an unfair shot. It is a model that has some capabilities that you just don't see in other places, which is good. And I think the website also has some features that you don't see in other places. And for the most part it's decently smooth. I'd love to see what they pursue in the future. I hope that they kind of can up the coherency, especially while keeping a hold of the good style transfer, because that would be totally awesome. That's the kind of thing I would be willing to pay a plan for in the broader scheme of things. I am getting dangerously close to just straight up not needing my Adobe subscription anymore and I think a lot of other people will start to feel that same way too. Especially the fact that Recraft can do the vector stuff. I mean that's pretty killer. There's a lot of tools in Photoshop that don't really need to exist anymore with AI image generation around. Like for example the Clone Stamp tool, which is still a good tool don't get me wrong, but I could just highlight a little thing and then just generate over it and it looks seamless for the most part. AI background removal is already getting really good and it's starting to get integrated into sites like Recraft. That's really the only thing like at least from for my workflows because I pretty much just use Photoshop for thumbnail editing. Like really basic work, obviously more advanced work. The creative control you get with stuff like Photoshop is pretty unmatched. Before you would need not just to pay for Photoshop, but also to have the skills and know how to use Photoshop to do something like turn my head into a watermelon or all these other crazy things. And now it can just be done in a matter of moments in various image generation websites. And obviously at the end of the day that's both a good and a bad thing. The downside is that there are going to be temporary job displacement for sure because of just overarching technology and AI. Why would someone pay a graphic designer in Photoshop when they can pretty easily do quite a lot of things in Ideogram or Recraft all on their own? Or just hire a freelancer for one quick job who uses AI. It's a lot cheaper and it makes more sense financially. Those are the short term job loss issues, but long term you get this democratization where the truly creative minds, the truly passionate minds are no longer limited by money, technical skill and know how and time. They can put their ideas out there a lot more quickly and a lot easier with with with few barriers. But that's how it goes with technological advancements. Out with the old and in with the new. Always a difficult shift. Let me know what you guys think down in the comments below. Thank you so much for watching today's video. It's just sort of like a recap review of Recraft. I really think it did deserve a video of its own. It's got some good features, and the people in my Discord server have been having a blast with it. I'll see you guys in the next video. And goodbye.