276 | The AI Automation Playbook: Turn One Idea into 1000 Assets (Without Technical Skills) with Ross Symons

0:00Ross Symons:Hello and welcome to Leveraging AI, the podcast that shares practical ethical ways to leverage AI to improve efficiency, grow your business and advance your career. This is Isar Maitis, your host, and I have a really great episode for you today. One of the biggest unlocks that you need to start learning as you start building things in AI is learning how to go from a single small experiment into an actually working solution that you can use and build around in a professional environment that will be repeatable and will generate the same results every single time. And in today's episodes, we are going to show you the mindsets, the concepts, the processes you need to think about once you start building AI automations in order to allow them to be scalable and actually support real business use cases.

0:54Ross Symons:So that's number one. But number two, one of the really magical capabilities of AI right now is being able to generate professional grade visual assets at scale. So images, videos, and so on for whatever need that you have in your business, but do this in a way that is, again, repeatable and automated as much as possible. So if you can combine these two things, you're getting the best of both worlds. If you know how to build automations well, it can apply to anything. and then also learning how to do this in a visual environment allows you to generate any visual assets you need for your business, and it does not require any previous technical skills.

1:36Ross Symons:So consider this episode of BOGO, where we are going to show you both how to build and think through building node-based automation processes, and at the same time, how to build really amazing professional-grade digital assets you can use in your business for marketing or for your website or for whatever it is that you need. Now, our guest today, Ross Simmons, is probably the perfect person to take us through this process. And why am I saying he's the perfect person? Well, he started his career as his developer, so he's very structured, well-organized, and well-planned when it comes to approaching business problems.

2:13Ross Symons:But he spent most of his career running creative companies and creative agencies, which means he has the perfect balance between left brain and round brain, which is not very common. It's actually pretty rare, but it's absolutely perfect for the topics that we're going to cover today. And since both of these aspects, knowing how to build automations and knowing how to build professional visual assets are really important and valuable capabilities, I'm really excited to welcome Ross to the show. Ross, welcome to Leveraging AI.

2:46Thank you, Sadi. Yes, good to be back again. Thanks. Second time, Ammonia. And yeah, Always a treat. Thanks. Yeah.

2:53Ross Symons:Thanks for having me. Thank you for coming back. It's, and I really think, you know, this, this topic is fascinating and timely and it's evolving. So I know we, we had node based automation tools and we had visual tools and now they're converging into this really unique set of solutions that allow to do things that were, well, impossible two years ago, difficult a year ago and now a monkey like me can do it so i'm very very curious to see what you're going to walk us through today i think it's like i said it's it's a it's a magic wand that once people know that they have can do very interesting things for their businesses yeah for sure i think that you know the um the node-based node-based i guess interfaces are not new they they have been around since animation and 3d rigging and i think there's some architect software as well that also does the same thing or uses the same interface and so some very smart person when they saw the ai sort of explosion happening they were like you know what let's just let's just give this a go and see what we can do there and i personally um from where i'm sitting i mean i'm trying to be on the sort of forefront of what's going on in the gen ai space particularly with the creative tools, you know, image and video and audio.

4:13And from where I am sitting, it looks like the node-based applications and the node-based interfaces are definitely the way forward. I mean, we train, as part of the company I run, we train creatives on how to use AI. And we take them through the basics of image generation and then how to get consistency across those images, then into video, but outside of a node-based application. And at the end of the course or at the end of the curriculum we show them Weavey or Flora in this case that's one of the tools I'm going to show you as well and you can see there's this thing that happens they're like it's all under one roof and everything you've shown us up until now and we've been you know multiple tabs open you've got to click there and then download this upload it there that is I don't want to say it's a thing of the past but it's when you see everything housed under that one kind of net on a single canvas, things just happen.

5:07And I mean, I can't pinpoint what it is, but it's just looking at the entire project in a single space and having all these little nodes and everything connected. Initially, when you do look at it, it's extremely intimidating because you see these blocks and it's like you think maybe it's this new AI sort of alien that alien interface that's now been put in front of you. But when you start explaining it to people and you start breaking it down to like each of these blocks represents a image or text block or a video block, and this is how you connect them. And when you start connecting the strings and pulling, you know, nodes across the canvas, that's when some really cool stuff happens.

5:51I mean to you know to what you mentioned earlier being a essentially a creative technologist but you know I think that someone that does appreciate the what it what what it I guess what a good interface what a good user interface and what you can do technically with something like that makes it more appealing I don't want to say it makes it more appealing to me but there are things that I know from a technical standpoint being a developer that I used to be um i can look at that and go oh i know how to add a variable in there i know how to iterate over that i know how to create an array to make and these are all coding terms but when you explain it to people on a very basic level and it's like cool well this is how you make multiple images from one image and yes the technical tools are still the same you're still using an array you're still providing one i guess single variable that that changes and something switches on so yeah i'm super excited about this space.

6:47I mean, and, you know, just to, I think the node-based, we touched on this earlier, but the node-based landscape is definitely, like, people see that it is a thing. They don't know why. And no one can really pinpoint why it is, because it doesn't really logically make sense to build this, like, spaghetti block interface to connect images and videos together. But when you see it working, and you see how scalable it is, I think that's another thing when you can see these i mean weavy refers to like that you're creating weavy machines or design machines and when you've got these blocks and you can duplicate that and just swap the swap the product swap the character swap the anything out hit the button you know drag all the nodes hit play and it just rolls it out for the rest of them you kind of see okay well this this

7:37Ross Symons:is very powerful stuff so yeah i think i think the big thing here is is is uh transparency it allows you to understand what's actually happening in a really complex process and i think that's why people are very easily connected to this because it allows you initially allows you to understand and it allows you to build right because once you understand you're like oh now I know what all these Legos can do and now the Legos make a lot more sense so previously like you said if you teach people oh you need to do this and then you need to copy this there and then download this file and then you grab the file and you upload this to the other tool and then it's like oh my god this is so complex but but but now so this it still gives you the Legos but you don't know really how to build with these Legos and and these these node-based tools that I'd like you to see on a single screen everything that's happening.

8:29Ross Symons:It's kind of like having the instructions for the Legos. Like, oh, now I know. And the other really cool thing is that all these tools, you know, Weavey for sure, before that Conf UI, which was very, very high tech geeky kind of like solution, but same kind of concept. There's a huge community, meaning more or less anything you want to build, somebody already built and sharing. You don't even like, oh, I want to build this really complex thing and it needs to do these 17 different steps. I'm like, oh, yeah, somebody built 16 out of those 17 in the initial process. You can add your one step into that.

9:05Ross Symons:And then in 10 minutes, you'll be up and running with a solution that, again, two years ago was not possible and six months ago was very complex. so let's let's really go into this and show and and as i mentioned i think it would be great seeing examples but also thinking you know one step back about the the design concepts that go behind it so other people don't just see the example but they can actually understand why it was built the way it was built and what are the benefit of of building it that way okay cool so So let's take a look at, let's look at this one first. And for those of you, by the way, who are just listening to this and not watching, A, you can watch this on our YouTube channel.

9:48Ross Symons:B, if you're on Spotify, you can watch this on Spotify. But if you don't, because you're driving your car or walking your dog or running on a treadmill or whatever it is that you're doing, we'll explain everything that's on the screen. So you can still stay with us on the podcast sound only, and we'll walk you through everything that we are seeing on our screen. Cool. So if you arrive at something that looks like this initially, you'd be like, what the hell am I looking at? Even when I look at these workflows, it's always still this like, okay, cool, I better just take it easy, take a breath and move slowly into it.

10:28because I think what's nice about this is you can look at the whole thing. It's not like looking at a piece of sheet music or looking at the parts of a car that are being put together. If you were a mechanic, it's very raw. It's kind of like, what the hell's going on here? But the more you zoom into it, it's almost like this world that you're kind of slowly zooming into. So like, and as you go into it, you start looking at the blocks. And when I'm looking at something like this, I think naturally the way these interfaces are designed. I mean, just as a quick example, you have a prompt box which you can pull.

11:08I mean, I'm not going to even try and explain how many of these little nodes there are because there's hundreds of them. But for the most part, you have a prompt box, you pull a string out of here into another box, and that could either be an image or another prompt box. And it could be, in this case here, is like an NELM, which is you get the option to select the large language model that you want to run the prompt that you put in. But the more you start using it, the more you realize it's pretty much the same as like a ChatGPT search with your little search button there. You've got the search block and that goes, you pull it out from the right-hand side of the block into another block which is to the right of it.

11:48So you're working from the left over to the right. So I think that naturally most of these workflows would start there. So it's not like they're just placed randomly and they're all over the place. There is some sort of structure. And only once you start using it regularly, do you realize that there is a bit of, I don't want to say magic, but there is a bit of structure and yeah, just, I guess, design thinking behind how you put all of this together.

12:16Ross Symons:Yeah, again, for those of you who are not watching, what we are looking at at the screen is there are three entries to a more complex process. Two of them are text boxes, which we'll explain what they do in a minute. And then there's an entry that is an image and they all go into a large language model as an entry point. So the way this works is that every one of these blocks can either be a input, a process or an output. and every one of these could be an image or text or video right so you can enter as an example in this case you can enter a prompt and a piece of code and an image into a large language model and because it knows what to do with these then it just does and then it can create whatever output you're asking it to create in the prompt now the other cool thing here because it's not necessarily one-to-one it could be one-to-many or many-to-one there are many cool things that you can do and i'm i don't know if ross is going to jump into this but just think about the fact they're saying well i don't know if gpt5 or claude 4.5 or whatever is going to give me a better solution for the thing that i'm trying to do you just drag another line from the prompt line and you connect it to another box and in one box you select claude 4.5 and in or 4.6 or whatever and in the other box, you select GPT 5.2 and you see the output of both of them on the same screen.

13:49Ross Symons:You don't have to go and open and copy and paste and have licenses. Like all of this goes away and you're like, oh, consistently when I'm doing this particular process, GPT 5.2 gives me a better output. I'm going to use GPT 5.2. I don't care about Claude for this particular thing moving forward. And the same thing with image generation and the same thing with Vigert. Like all these things are available with a very simple user interface. We can, you can just see the blocks on the screen and you can very quickly understand what each and every one of them does. And then I'll give it back to Ross to kind of walk you through one of these processes.

14:21Ross Symons:You understand how amazing this is. Yeah. So, I mean, looking at, this is quite a technical one in terms of it's got a system prompt as well as a standard prompt. I'm not going to dive too much into what that means. Basically, the system prompt gives the instructions as to what the large language model needs to do. The prompt box essentially is the variable data. So this system prompt will remain the same. The variable data you can swap out for multiple examples. In this case here, I'm basically just getting, I've created like what I've called an aspect ratio machine. So I can drop any image into this block here.

14:58That's going to be the reference image that comes in here to our large language model. The system prompt says, okay, cool. You are a JSON expert. JSON is the type of code. Don't worry about whether you know what JSON code is. It's just a different way of structuring a piece of text.

15:18Personally, what I find a lot easier to kind of iterate over and read because it's just easier to read. Anyway, it's quite a technical thing. And if you're a developer, you'll know exactly what it is. If you're not, just know that it's a different way of structuring information. So in the prompt box or the system prompt, I've got this thing that says it's all the rules it's like this is what you're gonna you're gonna create uh the crop must be x and no part of the image um must be regenerated no in painting allowed and i've given all these instructions bear in mind i did not write all these instructions i obviously used a large language model to write all of this but i was giving the outside of this i was giving chat gpt um just examples of like this is what i want to achieve help me create the system prompt so that when the system prompt does come in here, I can then just iterate over it and it's just easier for me to create.

16:09Then in the prompt box itself, I've said, okay, I want you to create different aspect ratios. So the standard, you know, one by one, four by five, three by four, 16 by nine. And the idea behind this is, you know, often we are presented particularly for social media. If you're creating social media content or banner content or anything digital, it requires a square image, potentially for whatever, or maybe it starts as a square image, but you're going to need a, for example, a YouTube aspect ratio, which is 16 by nine, and then a nine by 16 or four by five or two by three for the other channels.

16:44So your TikToks, your LinkedIn's, although LinkedIn has different structures or different aspect ratios. Anyway, so it's just, I just wanted something that I could just drop an image into it and illustrate the fact that you can turn that single image into multiple aspect ratios by keeping the image almost exactly the same. I say almost because these models are generative. They are generative AI models and no generative AI model, even by using the same prompt and the same seed, will generate exactly the same thing. There is always something slightly off, slightly different. And it is something to just bear in mind.

17:21I think that that's where the big difference comes in and where a lot of designers particularly struggle with this technology because they're like it's not exactly like i wanted it it's like you have to be a little bit forgiving or know how to comp what what it is you want to be exact back into

17:36Ross Symons:the image um yeah i'll say two things about this one is the models are getting much better at keeping consistency especially if you know how to give them the right references and prompt them two is the cost is so marginally low that you can create 40 versions of the thing until there's one image you actually like and it's still going to be almost instant and it's still going to be cheaper than actually going and doing a photo shoot of the thing that you want to get exactly or or paying a designer to go and take this image of this you know this this hessian bag or whatever it is this material bag and like please create put this into different scenarios that so that I can put it onto my newsletter I can put it onto my website I can put it across all five social media channels which all are different aspect ratios it becomes it becomes tedious and it's somebody's job maybe some people enjoy that but I for me I'm just like no there must be an easier way to do this so that's what kind of inspired this so running out of the LLM it goes into an array block and this is what I touched on about you know just understanding a bit about code it's basically just what an array does or it's just a list of a list of text items if you want to call it that each of those items runs out into a different box again I think these technical sort of aspects of Weavey specifically are going to get easier over time or maybe they stay like this I think the more you use it the more it just becomes second nature you kind of like okay well if I want multiple prompts coming out of this box then they are going to have to go into an array and then they're going to have to go into a list item of some kind.

19:14And each of these lists, you can then select which one you want. And each of those essentially is a prompt. So it's exactly the same prompt, but I've swapped. The only thing I've changed is I've just swapped out the aspect ratio for each of them. So they have multiple inputs here coming from the large language model. And then so running out of each of these, we have the list running into a, in this case, I've got Nano Banana 2, which is the latest craze. Everyone's using it. Everyone's talking about it. It's amazing. And it is cheaper and faster than Nano Banana 3 or Nano Banana where we know.

19:49Oh, whatever it was called, yeah. The one before. So Nano Banana 2, the one before was slightly more expensive and a bit slower. So, and just a snapshot here. You can look at, I mean, on close inspection, we've got a square version. We've got a 2x3. This one is a 4x5. There is a 9x16. Here we have our 16x9. and this is a wider one which is 21 by 21 by 9 which is like a lot more cinematic and wide but it's pretty much the same object i mean i didn't specify where within the frame it needs to be but i think if if you had to show this to somebody five years ago they'd be like how did you do that like honestly this is it might not look like much because it's only seven images but to create those seven images and and not only create those seven images but now have a machine where I can just go out here and I'm swapping the image right so I'd swap the image there which I ran all of these earlier so now I've placed a different image here what I'm looking at now is just a box of vegetables in a nicely sort of lit looks like a kitchen area and essentially just run all of those again and you've got at scale you know just by the click of a button took probably the best part of 20 seconds to run this again and we've got all of these over there and you can do whatever you want.

21:07Ross Symons:I want to add two very important things here. One to a little bit explain and the other is to share with the audience that is not watching this, what are we looking at? So really what we have is a system, a process that is repeatable, that the only thing that needs to change is the input image, right? You put in a new image, you get all the outputs, regardless of what they are. And I'm sure Ross will show us other examples in a minute, but all you need to change is the input and you get all the outputs that you built the machine to create in this particular case what the machine does is it creates multiple aspect ratios of the same thing now what you're not seeing going back to describing what's on the screen is both images had really unique backgrounds so there's one object in one of them it's a bag and in the other one it's like a basket of vegetables but in both cases the background has intricate details of shadows and countertop and it's not just a gray random background and in all images the ai manages to capture it and represent it and in conceptually invent because it doesn't have that information in the original image invent what's above below left or right of the original basket in order to be able to make it a 16 by 9 and so on.

22:30Ross Symons:The way it actually works is the prompt that then becomes multiple prompts asks it to describe the picture, recreate it, and recreate a new prompt that will describe the picture in the same level of detail as the original prompt, just in a new aspect ratio. And then what you give NanoBanana, you give it that new prompt together with the original image. So now Nanobanana is getting two things. It's getting the original reference image that it already knows how to reference, plus a very detailed prompt and a new aspect ratio. But you build all of this once. And now you replace the original image and you get whatever number of outputs that you want out of it.

23:14Ross Symons:And that's why this is so magical. And that's why these systems, if you understand how to build them, allow you to create, as I mentioned, resources at scale with practically zero effort after you've built it and troubleshooted. Yeah, totally. And I think the only real hurdle is understanding, firstly, how to kind of sit with the large language model and get it to a point where it's actually delivering consistent results because that's the thing that there's no one-click solution for. Yes, once the machine is built, it's going to reliably do pretty much what you've asked it to do every single time based on the image or multiple images that you put in.

23:58But it's really working with the LLM to kind of construct the prompt in a way that you are going to get consistent results. And you would think, OK, we'll just add as much detail as possible or, you know, surely can just reference the image. but it's not actually that because you have to, I mean, for me to get it to this point, I had to, you know, do it at least five or six times before it got some sort of consistency. I mean, I'm still not 100 % happy with, you know, some of the angles change quite, you know, just very slightly. Although it is the same, I mean, at face value, it looks like a photo shoot that was taken where they did maybe move the camera slightly, which I'm okay with.

24:39And, you know, it is, you have to be a little bit forgiving in terms of what it is but for the most part this is in in my opinion miraculous technology it really is i'm with you 100 i i would add one more thing

24:52Ross Symons:and then maybe we'll dive into another example is going back to what i said in the beginning don't be intimidated with knowing what does that mean what ross said right now how do i get to that first starting point copy somebody else yep all these platforms there are hundreds of thousands of examples that people are sharing the entire process with their prompts, with their connectors, with their arrays, with like everything. Go and search for, I need a Weavey that does one, two, three, four. And there's a very, very high chance you'll get something that either does exactly what you needed to do or 80 % there.

25:27Ross Symons:And that's also the best way to learn because now you can dissect what this thing is doing. You can see exactly how the prompt is working. It's like, oh, all I need to do is change this one thing in the prompt to make it mine and you solve the problem so don't yeah don't be afraid of staring at a blank sheet and i don't even know how to get started start with three or four that already are there because the community is sharing it and then reverse engineer from that yeah exactly that term you've just used now i think that that is what ai is i think in my opinion like solely built for is to reverse engineer ideas we like we look at all these things now and you know certain tools make it easier but you look at this concept you're like we want to create an ad campaign that looks like that that that and that mixed together we know what we want it to look like how do we work backwards through the machine to get our raw materials like we've got these brand assets we've got images we've got brand identity or corporate identity how do we plug it into a machine that spits out what we have in the format that we want.

26:32And that's, I mean, that's the simple computer process, input, process, output. It's no different. But what AI allows us to do is now gather a lot of reference material and a lot of ideas and reverse engineer, like break it down and break this image down, break this video down, break this concept down into its simplest form so that a large language model can understand what it is and take those key points, feed it back into the system so that you've got the raw data. And every time you send it back into the system, it produces exactly the same result. And I think that we've never had technology like this before.

27:06And that's, to me, what is very exciting. A hundred percent. Okay. So let's take a look at, I've got another one, which is quite a similar one. This one I've actually shared on LinkedIn. This is an emotion machine. So very similar concept, except we're not putting an object or a product photo in here. We're putting a person. So it's the face of a person. So here we have these images I created in Mid Journey. So it was just a portrait shot of a woman. She's got dark hair, kind of gray streaks in her hair, some freckles, and she's looking directly at the camera with a pretty plain look on her face.

27:44Same idea. We've got our system prompt and we've got our JSON prompt, which has got all the expressions. So I've got nine different expressions. Here we have laughing joyfully with head tilted back. Something just in terms of the language being used here, I make this as generic as possible so that I can put he, she, her, it, they, it doesn't matter what I put in there or which object or which subject goes into the reference image block. It will always produce the same result. So, yeah, looking thoughtfully with a hand on her, like I said, her in there, that example, but it didn't really affect it.

28:16Pouting sadly and looking down, winking playfully with a slight smile. So, these are my nine emotions that I have. Cool. They run into the large language model, split out into an array. Each of those gets spat out into a list item. Each of these gets selected, pulled into a nano banana block. And now we've got exactly the same person, but with completely different emotions. Essentially, this is a photo shoot happening right in front of us now. So this is absolutely brilliant.

28:44Ross Symons:So I play with Weevy a lot. Not play. I use Weevy a lot. And play. It's a really fun tool, right? If you have any creative ideas, it's just a fun tool to use. But I've never tried to do emotions with it. And again, for those of you who are not watching this, this is freaking incredible. It literally looks exactly like the original person and the emotions look fantastic. And it's like you're saying, it's like the perfect photo shoot because now you have the person in multiple angles and multiple expressions. some are funny some are sad some are deep thinking it's just incredibly good yeah it's amazing like honestly i didn't i didn't think that the results were going to be as good as they were and i thought okay cool i got lucky with the first round and exactly the same as we've done in the previous example like swap the character out so i grabbed this guy he's a um he's an asian guy he's got like a tattoo on his face sort of blondish hair bit of a goatee and i was like okay cool Let's see how close it gets to that.