0:12 Hello and greetings, our viewers. Welcome to Ethio AI Insights. This is Behailu. In today's program, we will look at using Artificial Intelligence to generate various images, especially mainly using Seedream 3.0, the newly released image generator AI, that is. And also, using ChatGPT's 4o, we will generate various images. Then, next, using FramePack, we will see how we convert these images we generated, or other images, into videos. Both tasks we will do will be solely using Artificial Intelligence. So, if you are new to the channel, start by subscribing. Also, liking, commenting, and sharing are encouraged. We will put the prompts we use on our Telegram page. So, join our Telegram page.
0:58 Stay with us until the end of our program. Currently, in the world, among all the Artificial Intelligence models, the one ranked number one in creating the best picture or image is OpenAI's GPT-4o. However, now another new AI that creates images has emerged. This is ByteDance's Seedream 3.0, so to speak. Based on various metrics, the two are almost at an equal level by different criteria. Using both, we will create and compare various images by giving different prompts. So, we will see which one is good, which one is better for us. Accessing it is very easy. First, where do you go? You go to Dreamina. Dreamina, as you can see above, this is their website. If you go to dreamina.capcut.com/ai, it will take you directly here.
1:49 To the AI called Dreamina, that is. Then what do we do? We get something like this. They do many kinds of things here. As you can see, there is an image generator. There is a video generator. There is an AI avatar generator. That means they have many things. But what we want now is the image generator, so here is the image generator. We can click the 'Generate' button here. Alternatively, we come to this side, on the left side, that is, and click on 'Image generator'. So, I will click this one. When we click this, it comes up like this. So, when this comes up, these are things I generated before. As you can see here. So, after doing that, when we come here, there is something called Model. As you can see here, the Model... under Model, it says Image 3.0, and then by Seedream 3.0.
2:37 So, we can select that one. They also have other things. For example, you can adjust here. They have various models, previous models, that is. We don't want those now. The latest model is Image 3.0. It's new, as you can see here, it says 'New'. It was released two days ago. So, that's what we want now. Regarding the resolution, we can do 1K resolution or we can do 2K resolution, that is. The only difference is, down here, as you see, there is Credit. If we make it 2K resolution, it will cost three credits. If it's... uh... 1K, then it costs 2 credits. As we can see above, there are credits here. So here, you have 120 credits every day. So, this 120, every single day, they give you 120 credits. So, if you use it... you use it. If you don't use it, it expires.
3:25 So, daily, they give you 120 credits. Let's give it a prompt. We'll give it a prompt here. As you can see here, a prompt that serves for marketing purposes, I have given it one prompt here. We will give it this. We give it this and then we will see how it generates it. Here, the aspect ratio, I come to Dimension and make it 1 by 1. To help with comparison, that is. Since ChatGPT also uses the same aspect ratio. So, I come here and, as you see, I have given my prompt here. Then I click 'Generate' here. So, it's generating. It says 'Video' here. I come here and go to Image and look here. It shows it under Image. You have to select here. Then, as you can see here, we are working on it.
4:10 As you can see here, it is generating. I have given the same thing to ChatGPT. I'll show you on the side. While this is generating, I gave the same thing to ChatGPT, and as you can see here, it created this for us. It created this for us. As you can see here, GPT-4o, currently in our world, at a very high level, creating such clear things like this is done by GPT-4o. It's OpenAI's, that is. So, as you can see here, the work it gave me for this prompt looks like this. It's extremely beautiful. It did a very nice job. It generated a very neat thing for me. So, after looking at this here, we come back to this one. This one is still working. So, we go back to this one and look again. As you can see here, the... the can itself, when you look here, it's very nice. The lemons here, you see, the ice... maybe a little bit...
4:59 the... the ice, there's a slight watery quality here, we don't see it when we look at the background behind it. Other than that, it gave a very, very excellent result. Will Dreamina do this? Will Seedream 3.0 do this? When we look at that, as you can see here, it's working. As you can see here, it's working. It gives four at once. Maybe compared to ChatGPT's image generator, what makes it different here is it gives four options. It gives four different kinds of options. When we come to ChatGPT, though, it gives only one. But regarding quality, ChatGPT is extremely... here... We can see here that it produces a very clear result. As we saw in our introduction, that's the reason. Measured by various parameters, ChatGPT, in terms of image generation , is the better one.
5:48 Ranked second is Seedream, as we saw earlier. So, it is generating here. It took some time. It used to be very fast. I don't know why. Well, as we can see, it has generated something like this for us. It's a very nice thing it generated here. As you can see, since it showed all four, it generated four options. Here it gave us four choices. As you can see, it's an extremely beautiful, extremely effective image generator, Which one will be better? Regarding that, you can use different parameters and do things.
6:36 In terms of quality here, though, as far as I can see, apart from giving one image, ChatGPT did a good job here. Dreamina, well, here it also did a very good job, as we can observe here. This is the first one. Let's start the second challenge. On the second one, as we look here, we are on ChatGPT. We are on GPT-4o. What is the main purpose of this one? It's not just to generate images, but does it understand the images or texts within the image? Does it render the texts correctly? That's the question. Many previous Artificial Intelligence models, image generating ones, could not correctly identify text. But now, can the current GPT-4o identify this?
7:21 Similarly, we will go to Dreamina and see if it can do this thing. So, on GPT-4o here, for the prompt I gave it here, as you can see, this is an extensive prompt here, as you see. For this prompt, it gave me this. It found it in a very surprising way. As you can see here, it placed the texts here correctly. This is on a diary where these texts are written. It picked up these texts precisely, every single word here, correctly, as we can observe here. These prompts, as I said, we will share on Telegram. Join our Telegram channel. This prompt, not just this prompt, but in the future, we will share various resources there. So, well, this one, in a very quality way here, the GPT-4o image generation here, as we saw, it worked like this.
8:10 Now let's look at the other one, at Seedream 3.0. Well, here for the same prompt, I got this here. On Dreamina, what I get looks like this. As you can see, Seedream here 3.0 gave me this. Here it gave four options. The first one here, as you see, it did a good job with the... the leather item, the diary being a leather product, etc., it found that very well. However, the text has a slightly blurry quality, it found the top part correctly. Inside here too, it found it correctly, but around this area, there's a bit of a smudged appearance. That part needs a little fixing. This is what the first version gave me. The second version is also similar here, as you can see, the second try... it gave four at once. We see the same thing on the second one as well. The third one here, it's inverted. On the inverted one, it looks slightly better,
9:00 but because it's inverted, it's not very easy to read correctly. However, if you look closely, it seems better than the previous ones, as we can see here. Looking at the last one here, it did a good job. Here, it seems better than all the others, but still here, there are some missing spots. Compared to this one, with GPT-4o, well, GPT-4o's work is very clean, as we can observe. However, here too, Seedream, when we look at it, it also did a good job, as we can see here. Well, now, another prompt I gave it here, as you see, "An Ethiopian girl riding a bike with flowers in the basket on a sunny day :)" I gave it that.
10:18 As you can see, this one here, when we look, it created another one here again. As you can see, it did a very nice job. Well, GPT-4o, what it did here, when we look, well, the one riding, the bike riding one is there.
11:08 It's very pure. The only difference from the other one is that here it only gave one image. That other one gave four images. So, both here, at an excellent level, we can see that both of them worked very well. So, the prompt I am giving now is "A teacher writing on a blackboard in a classroom with attentive students, natural lighting Here, it looks like a rural area classroom. What it created here. Let's see here... Well, here it made a female teacher, which is very nice. It gave four different options like this.
11:56 Now, what we will look at is converting the images we generated into video. Among the videos we will convert, one is this image generated on ChatGPT. To convert images on our computer into short videos, we need to install an AI model called FramePack on our computer. Where do we find this? When we look, as you can see here, we find it on GitHub. We go to the GitHub address highlighted here, this GitHub address, we go there and then what do we do? We read it. First, what are the requirements? We look at that. Here, as we see, FramePack can generate video. However, here we must note, they are only on GitHub. They don't have any other web applications. But there are many web applications released under the FramePack name.
12:43 All these web applications are not associated with this AI, they have explicitly stated here. So, any websites that ask for payment, saying "generate for a fee" related to FramePack, don't use them. Because here, as you can see, it indicates that they are spam and fake. Other than that, FramePack is Apache 2.0 license, meaning it can be used for commercial purposes. You can use it as you wish. Once it's installed on your computer, for free, whether you have internet or not, it works locally.
13:35 Regarding installation, when we look here, it's not complicated at all. It's extremely easy. But there are things we need to look at here. There are requirements. it needs Nvidia GPU. Other than that, at least 6GB GPU memory is required on your computer.
14:21 Other than that, the installation process is extremely easy. As you can see here, this part is for Windows. You just need to click this link here. You come here and click this. Then, after you download it, unzip it and store it in the location you want. When you open it, it looks like this inside here. This is what it contains inside here. System, webui, environment, run, update. Just like this, something like this will appear.
15:11 Here, the first thing you need to do after installing is to update. After that, you can click the 'run' file. Then you click 'run', what appears is this. Then first, something like this appears. This is running on my end. When it runs, it looks like this. Then, on the side, it opens this. So, this is, well, it's on the local environment. It's not from anywhere else. Here, as you can see, this one, for example, I gave it to convert this image to video.
16:41 It does it in reverse. It starts with the finished thing first. Then it goes in reverse. Don't get confused there. So, you wait until it finishes. Well, here there are 25 processes. As you see here, Steps, there are 25 steps here. It does all 25 steps, and at the end, it gives you the video. So, from other things here, Total Video Length, you can change this up to 120. Here I set it to 5, but it can go up to 120. The longer the time you increase, the longer it takes to generate, the time it takes increases. That's understandable.
18:36 So, keep that in mind. Well, first, the energy drink image is what we are converting to video. As you can see here, well, as you see here, how FramePack converted it into video, we will observe.
19:38 Well, viewers, using only Artificial Intelligence, we have seen how we generate images, and videos. In doing so, we looked at GPT-4o, FramePack and also Seedream. So, this is the program we prepared for you today. Subscribe to the channel, like, comment, share. Until we meet again next time with another video, have a great time.
