181 | How to Ideate, Generate, and Animate Scroll Stopping Images and Videos with Arminas Valunas

0:00Hello and welcome to another live episode of the Leveraging AI podcast. the podcast that shares practical, ethical ways to improve efficiency, grow your business, and advance your career. This is Isar Maitis, your host, and we have a really exciting show for you today. As you know, creating visuals that gets people's attention has been a key component of everything that we do, whether you're in marketing or you just like to share content online or build networks. Being able to capture people's attention by generating the right images or the right videos has been something that companies has been born and grown on.

0:38And if you know how to do this right, you can grow your business, whether you are a part of a big machine, like running your role in a marketing department or a sales department or something, whether you're a solopreneur or a consultant, a content creator, all of the above, you need to know how to create visuals that attract people's attention and connect with the people that you're trying to connect to. And it needs to be aligned with your brand and it needs to be aligned with your tone the voice and it needs to be a lot of things in order to actually do it right. Now, it used to be a lot of work, especially when you go to the video side of things, you needed to actually have a good story and a good idea.

1:11And you need to be able to create the right process with videographers and camera people and audio and visuals and lighting and actors in a studio and all these things to actually get these things done. And the good news is now you don't have to. You can create incredible visuals without having a camera and without knowing anything about design. Now, obviously, if you know about design, you can do a lot more and a lot better. We're going to talk about this in a minute, but you don't have to. And you can actually use AI tools to do this. Now, another good piece of news is now there's a few tools that aggregate a lot of these capabilities into one tool where you can do most of the steps of the process in one place without leaving the tool.

1:54And this tool brings into it multiple other tools and AI resources that otherwise you would need to jump back and forth in order to create the output. So today we are going to learn how to create amazing visuals that will be thumb stoppers on social or will grab people's attention if it's an ad somewhere or on your website and how to do this for both images and video. And we are very lucky because we have Armin Valunas with us. And Armin has been doing this for a very long time. he's been in design as well as running a company that does branding and design for other businesses for many, many years.

2:29So he both understands the visual side and the concepts and the professional side of this. But in the last couple of years, he's built all in on AI and he's been posting these incredible things on LinkedIn. That's how I found him. And he's teaching courses on how to do that. So he really brings to the table everything you want as a guest of this podcast. He has the professional background. He has the business background. He's been running his own business for the past six years, and he knows AI very, very well, and he knows how to teach it because he's been teaching courses. So if you want to learn how to create incredible visuals with AI, don't go anywhere in the next 45 minutes to an hour, or if you're driving, stay put.

3:07This is going to be absolutely fantastic. I'm very, very excited to have you. Armin, welcome to Leveraging AI. In the next few years, AI technology will change our world dramatically. Whether you are a business executive trying to catapult your business forward, or just somebody who refuses to be left behind and want to advance your career, this is the show for you. I'm your host, Isar Maitis, a serial entrepreneur and an AI enthusiast. You'll hear invaluable practical tips from innovative business leaders, AI practitioners, and some of the brightest AI minds in our world today on how you can leverage AI in ethical ways to advance your career and grow your business.

3:55Thanks, Saras. Thank you so much for having me. And yeah, I think that will be a great workshop and I have prepared a lot for you guys. I want to mention that, of course, the experience in advertising industry is so important. And I have been working with the biggest brands like IKEA and Swedbank and others. But at the same time, I want to encourage you, like Isar mentioned, these tools, AI empowers you to create and just throw yourself fully into AI and you will be able to create the most capturing and the highest quality visuals. So I think let's jump right into this, right, Isar? Awesome. Yeah.

4:36Before we jump in, just for those of you who are joining us live, either on LinkedIn or on Zoom, thank you for joining us. I know you can do other things on Thursdays on noontime Eastern, but I know a lot of you have been joining us every single week. So I appreciate every single one of you. I also really, really want to thank you for participating. So if you have questions, then ask them in the chat. Again, whether you're on LinkedIn, you can just say it in the chat. If you're on Zoom, you can say it in the chat as well. For now, just go ahead and introduce yourselves. This is one of the reasons you joined these lives so you can network and get to know other people.

5:10So tell people where you are, where you're from, what you're doing, what you're looking to get out of this. But like I said, if you have any questions, please ask them. So that's if you're here with us live. If you're not here with us live, the question is why? We do this every single Thursday with amazing people like Armin that are going to share very practical use cases with AI. And then you can ask them questions if you're around. And if you're not, you can still listen and benefit from it, but you can participate and you can engage. And you don't get to experience that when you're just listening, but there's a lot of chat happening in the chat as well.

5:38So you can join that conversation if you join us live. I will say one last thing before we dive into today's topic. Today's show is brought to you by the AI Business Transformation course. It's a course I've been teaching personally for the last two years, at least once a month. So we've taught hundreds or maybe thousands of business professionals on how to implement AI in your business. The course is four sessions of two hours over four weeks, and it takes you from whatever your level of knowledge is right now to understanding all the tools, what they are, what they can do, how to use them. A lot of hands-on experimentation across multiple aspects of the business.

6:15And we end with business strategy, an actual blueprint on how to implement AI successfully in your business. If you haven't done anything like this, anything structured on how to implement AI successfully, you owe this to yourself, you owe this to your business, and you owe this to your career. And so for being a listener of this podcast, you can come and join us with a promo code. And the promo code is leveraging AI100, all uppercase. There's going to be a link in the show notes, and we're going to share a link in the chat right now for those of you who are with us live. And you can come and join us for the session.

6:50The next session starts on May 12th. So if we do these courses, like I said, every month, but most of them are private courses. We teach them to companies and organizations who book us to do it just for them. And we open a public course once a quarter. The previous one was January and February. Now we're opening one in May. So if you need that, and if you haven't done it, you need that. Come and join us on May 12th. That's it. Now to our session with Armin. Armin, the stage is yours. Show us your magic. Oh yeah. Thank you, Sar. Thanks so much. So I think I will start from sharing the screen and yeah, we will go through, let's say, let's check the last version of video that I've done.

7:31And we will try to recreate this as fast as we can. That would be fantastic. For those of you who are not watching, by the way, if you're driving your car and so on, that's perfectly fine. We will explain everything that's on the screen. But if you can, there is a YouTube version of this if you're watching this after the fact. And in your show notes, you can just open it and click and go and watch the YouTube video. And then you can see the visuals as well. So let's watch this for now. Wax London, ethically crafted pieces designed not just for seasons, but for life stories. The foundation of lasting style lies in superior fiber.

8:12Wax London, Clovis with character. So for those of you who can't watch, this was incredible. And I still can't believe this is AI generated. So it's different images and not images like videos of a fashion show and then actual people in different scenarios like driving a boat and walking in a beautiful old city in Europe, wearing nice fashion clothes and bracelets and so on. And I'm blown away that this is all AI. So I'm personally really excited to see how this was done. Yeah. And it was so much fun creating this. So I think I will go to the whole process and we will see how to do this. So at first, really big thanks for my colleague, Joe, who inspired me to create this.

8:59Like I should mention that this is real product, real garment. I took this from Vax London page and I just downloaded these photos, like you see. And I trained Laura. And then I trained the second Laura for character for my appearance. but I didn't use this in the video. And I generated all these first like static images and from them like videos and then created the whole showreel. So let's dive deep into the whole production process. So I chose the FreePic and it actually gave me like really high quality. And at the same time, I think it's really like user friendly and really easy to produce these images and later on all the videos so here we are in the home page of 3pik and from here if we want to create the images we are going to create from here we want to train the object the garment so from here we are choosing object and just have to press plus button but actually i already trained So let's go through the process.

10:08So here we say the name, let's say wax, then shirt, the type. And here we upload all the images. I will upload these images. While you're doing this, for those of you who are not watching, FreePick is one of those aggregators of tools that brings you a lot of different other tools into them and make it very user-friendly to use. And what we're doing right now, when Armin says, Laura. Laura is just a way to train a visual model on a specific person or style or object. You can train it on different things. And when you train it, it can then knows how to regenerate new images of the thing. In this particular case, a high-end fashion shirt.

10:47So what Armin did is he basically just opened and said, okay, let's create a new training, uploaded multiple images that he, when I say multiple, it's seven, like it's not 50. And he has seven images that he downloaded from the website and is basically going to create a new model that knows how to create this shirt in any scenario on any person with any background because it learns how to create the shirt. This is exactly what I have done. And actually, to be honest, you need to upload. And my kind of recommendation would be all the time, follow all these tips. And it says 12 to 25 images. So for this if you don't have 12 at least because if you have only seven it will not allow you to train the laura but then what i did i basically changed the composition of the images and flipped the images horizontally and i have 12 images and i was able to train this laura and yeah actually it's so on point i want to interrupt myself a bit and maybe to explain how i understand lauras there are so many technical terms about lauras but in general it's a low rank adaptation and let's say this is our like flux and by the way we are training on flux so let's say this is all the flux model and all the whole data is in here so we are training like just one little piece like this is our like train laura and it shows the model like if it's on a object or like character it shows how we want to generate the special, like this special object.

12:21So in this case, I already generate Laura. So let's say we generate it because otherwise we would be waiting a pretty long time. Actually, it's quite fast, but maybe it still takes 15, 20 minutes. So we don't want to do this. So I created Laura already. We'll do the cooking show thing, right? We're pulling out of the oven. For those of you who don't know what Flux is, so there are several different leading visual image generating models in the world today. Probably the best open source one is Flux. So it's just a name of an open source model that works very well. Because it's open source, there's a very easy process on how to train it.

13:01And in this particular case, it's built into a simple user interface that runs in FreePic, which makes it even easier. I think most adaptable, and if I can say this, like trainable models out there because it's open source and you can do whatever you want with Flux and it gives you such a high quality. So definitely I recommend if it's for training objects, characters or styles, just use Flux. You can do this, by the way, on like older ones like SDXL or SD1.5, but Flux is the best at this point. We have already our trained object, in this case, this Vax London shirt. And by the way, there are many pre-trained objects in FreePick that you can just choose and maybe test if you don't want to, let's say, train your own lore because it costs a lot of credits.

13:50So at first, it's a good idea to play around with the pre-trained object. So when you have this train object, in this case, our garment, we just select this one and at Lidovax, it's my train Laura name. And it appeared in this field, in this box. And here, if it's active, it will generate this train Laura product. Here I can start typing and creating prompt. But actually, I already have a lot of prompts prepared for us. So I will not be doing creating from scratch. Let's dive deeper a bit into my thinking. So I have a product and I was exploring two cases, like consistent character and different characters.

14:33And yeah, let's talk a bit before I start generate. Let's talk about the process and the thinking. I divide, or at least I am working in three stages. So the first stage is pre-production where like creative ideas are born, where I'm creating like mood board where I find the tone of, not tone of voice in this case, but the color palette and the whole vibe and mood. And then I am usually drawing some sketches like rough sketches, but if you don't want to draw, it's not necessary, but just creative process and the pre-production is so crucial and so important for getting the best results. And also, of course, nowadays helps like AI tools and LLMs in general, let's say ChatGPT, it helps in every process, especially in creative process and pre-production.

15:21So I'm like chatting with the, like generating creative ideas, chatting with the, in this case, ChatGPT, but also I'm exploring other tools as well. So I think it really depends on you and your style, but nowadays it's just the leveraging your creative capabilities and you don't have to let the AI kind of drive your creative ideas and your creativity, but you should empower and leverage your creativity with the AI. So the second stage is production itself, where I am generating images, videos, making some selection from all the material that I have, editing, combining music, voiceover, and everything falls into place.

16:05and last touches. Also, this is really important. I think this is where magic happens is post-production. And it's really important because retouching and color grading and it just lifts you into this like pro quality that you are reaching. Yeah, so quick summary, three steps. One is pre-production where you get the ideas basically and more zoom in and what exactly you're trying to create. And as Armina said, if you're very creative, use your own creative ideas. If you're not, you can use large language models to help you through the creative process and think of an idea that will work for your target audience that is aligned with what you're trying to do.

16:43Literally just have a conversation with it, explain what you're trying to do, and it can help you. Creating mood boards, rough ideas on how the video. So the video that we've seen had like somebody, it started with somebody walking in a fashion show and then people walking the street and then somebody in a boat and then a close up. There's all these components that in your head, you need to know that you need in order to actually show them afterwards in the results. So this is more in the creative process. And then you create all these components and then you put them together. So now let's go to the steps of actually how to do it.

17:13Let's do this. I was jumping around a bit, but it's really important to mention these steps because if you skip something, it will be messy. And it's even like before AI, all the time we are using this kind of process. Let's say we already have trained object, garment in this case. And let's say I'm just create like some simple prompt and then I will compare how it appears with the advanced prompts. So man working in the street, natural daylight. It's important to point that this works with a flux model. And if you want to, oh, it's, and you see it's not working all the time. So it gave me like four variations, but the first one is maybe the most accurate.

17:58That's why I'm more into other tools and like using ConfUI, but this is a whole another topic. So it's important to mention that here you can demode and you can choose the model. So it's like you see it's Mystic, Etagram, Imagine 3, but it works only with the Flux because we train Laura on Flux. We can't train on like other models. Yeah, so again, for those of you who are not watching, in FreePick and similar tools as well, like CREA, you can choose many of the image generation models. But now because we trained Flux, we got to use Flux. But if you're just generating images, not on a trained model, you can use all of them.

18:38Exactly. I'm using a bit more advanced prompts. And this is also a deep topic. It is an art of itself. And when I'm using more advanced prompts, it gives, well, for me at least, it is much better quality. and much more interesting. Let's read a section of this prompt just to tell people the understanding so they understand what you mean by a more advanced prompt. So let's just read some of it. So the prompt, length, fashion, runway, short subject, focus, male model, walking, wearing white shirt, subject details, fit male model, white swept back, brown hair. We point more accurate details to everything, to characteristics, to character, what he's wearing.

19:20And when you're prompting with the loris in this case you shouldn't be confusing the model let's say if i would write like red t-shirt or something totally different it would be confused and won't give you like good results but this is also maybe i want to show you a bit where you can start like how can you learn how to prompt so i think the best start is just let's say download this one and just to upload the image into the chat GPT and just ask, describe the photo. And I'm in the camp where like people are saying like, but you are copying, you are doing this or that, and this is bad. But actually it's not like you are just trying to figure this out.

20:05What model, because vision model itself, it gives you the answer, how it interprets this photo. And you just borrow these key phrases, but you are making your own prompt. It's a really good base. Again, to explain for those of you who are listening, you can take an image that you want to mimic the style of or the concept, or in this case, it's a model on a runway. How to start prompting better? You can just drop an image that is your inspiration and say, I want to write a prompt that will generate a similar image in an image generation model. I want you to write a very detailed prompt that will mimic everything in this image, including the style, what he's wearing, the background, and so on, the lighting.

20:43and it will give you a very long detailed image that then you can use as your baseline to make changes and massage to whatever you need. One more thing with regards to prompting retrained models with Laura is, I'll say more extreme than what you said, it's not just don't confuse it, but tell it what it is. If you would say a buttoned up stylish white shirt, describe the thing that you trained it on, it makes it even better and it will give you more consistent results. So don't just trust the training of it to be perfect, but also describe it in your prompt and you will get more consistent results than if you don't.

21:17So let's continue with the process. So now you have the images of the guy in the fashion show. I assume you do the same exact thing for all the other characters, right? The guy in the boat, the guy in the street. I'm actually very curious about the close-up one because one of them was very different, right? Most of them were like full body, relatively kind of like style was similar. One of them, you literally specifically asked for close-ups of the fabric and close-ups of the person. tips and tricks and i will reveal all the secrets okay maybe let's dive deeper and maybe let's go with the like generating video and then i will go to this macro shot let's say we really love the shot then what usually i'm doing actually you can go like straight to the great video but actually i'm upscaling and it's really good that freepic has the magnificent upscaler which is the best upscaler, I would say.

22:08So I'm editing. I'm pressing like edit button and then I'm upscaling. And it's basically the same parameters that in native Magnific on its own. But yeah, everything is the same. I have a few presets. When I don't want to change the character appearance, I tend to lower all the creativity stuff parameters or like portrait soft is really good. So let's say I will upscale and it will take some time, but not that much. During this time while we're waiting, I can reveal my secret. So actually it's not about generating, especially when we are working with the real products, right? And garments and clothing.

22:50It's not all the time just generate for the sake of generate, at least in my thinking. So I actually just took this photo. It's a distinct photo. But I took this and actually I generated using runway. Runway is really cool. And especially like John Ford, they released this latest model, I think two weeks ago. So it's really new. Runway version 4. So that's a good trick again for people. For those of you again who are listening, what he's done to get the really extreme close-ups is actually use actual close-up photos from the website and then animated them with tools in this particular case. Runway version 4 is really good with consistency, which is very important when you're trying to do stuff like that and actually have the shirt look like the shirt.

23:32So now you have the standalone components, you upscale them, then you animate straight from there? Yeah, actually from upscale photo. But when we upscaled, we can't generate video directly from here. So we have to go back to my creations and press here. And now we're talking. So now we can create video. So we are pressing create video. And here is the video. And also you see the section, the first stage images and second section would be videos. It automatically gave me a prompt. But in this case, actually, it's quite good. A male model confidently walks down a minimalist runway. So actually, in this case, I will use this.

24:16But I can give away another secret. What I'm doing, I created my own Armada video prompter. And usually I'm just using this. It just elaborates your prompts. it just makes better. So I upload the photo and the man is walking towards camera. And we will get really detailed prompt and even two prompts actually, where it describes what is the composition, the wide angle shot, how the model looks like. And all these things helps to generate like much more accurate videos. And actually it made my videos much better. I just want to pause you again to explain, for those of you who are not great prompters, and especially when you're moving, even if you are a great prompter on an LLM, and now you need to prompt a video model, it's a little different.

25:06And the best way to do this is to have some kind of a framework. And if you don't have one, in this particular case, Armin created a custom GPT that does it. And all he does is he uploads the image that he's already created into the custom GPT and tells it a general idea of what it is. And it creates a very detailed prompt. Now, if you don't know what to put in a video prompt, you can ask ChatGPT. Say, I want to build a video prompt. What component needs to be in it? And it's going to tell you it needs the composition and the lighting and the motion of the camera and the motion of the character and the feeling you want to get and stuff like that.

25:39And then more details and different characteristics of the person and what's happening in the shot. And so it will tell you all the different components. And then you can ask it to build like a prompt generator for you based on that if you feed it an image. and then he will do the work for you. So it sounds really fancy, but you can start very quickly and get probably not the same level of results that Armin got because he's been doing this for X number of years. So he knows what he's looking for, but he will get you from knowing nothing about it to 75 % there, which is way ahead of most of the people in the population.

26:12Yeah, definitely. And I want to share some thinking behind this. So actually you can do this perfect custom GPTs on your own. Just let's say this is a great example. I've created also a custom GPT for generating, describing pictures. So you just go to the page of the model itself. Let's say this is frames by runway. This is like prompting guide. So the best thing is not to try to guess how to create a custom GPT. These developers are giving you all the information, how to prompt. Just copy this, all the info with the example and feed it to the knowledge base. and you have perfect custom GPT. Great idea.

26:52Yeah, let's continue. I decided my own prompt with my custom GPT is better. So I copy paste here and let's check a bit about the video models itself. So now it says Cling 1.6 Pro, but you can press here and you can choose all different models. So I think in my experience, it's still Cling 1.6 is the model that I am going most often with, especially when using FreePick. Like Google Geo too, it's also really super good and like really high quality. But I'm using more from text to video because actually from image to video, it is just in my experience, like Cling delivers better results. Yeah. And again, for all of you, one of the beauties of tools like FreePick is that it gives you access to all of these tools, right?

27:45So you can pick and choose is whether you want to use this image generator or this video generator. And the cool thing is you can try it with three different ones. And yes, you're going to pay tokens for all of them. So it's going to cost you a little bit of money. But if you're doing this for a business purpose and not just for fun, then it's still going to be significantly less money than you spend ever on creating a professional video for anything. Yes, you're going to run this on Vio and Kling and Sora and Runway. And all four videos are going to cost you$10. It's still nothing compared to what it would have cost you to even just get the actor for 30 seconds.