EP 3: A Guide to Generative AI Tools for a 10X Increase in Creative Content Output

Transcript for this Episode:

INTRODUCTION VOICEOVER:
This is Tech UNMUTED. The podcast of modern collaboration – where we tell the stories of how collaboration tools enable businesses to be more efficient and connected. With your hosts, George Schoenstein and Santi Cuellar. Welcome to Tech UNMUTED.

GEORGE:
Welcome to the latest episode of Tech UNMUTED. I'm George Schoenstein, your host.

SANTI:
And I'm Santi Cuellar, your co-host.

GEORGE:
And we are joined today by Sam Husbands. I worked with Sam for a number of years - he's our head of digital and demand here at Fusion Connect. We're going to talk today, about a bunch of different things from an AI perspective, in particular, Sam's been working with Midjourney for some image development and some other elements that we use from a marketing standpoint. But I did want to sort of open it up initially with a quick discussion with Santi about some of the things we've been doing from a bot standpoint internally, you wanted to give us just a quick view of that Santi?

SANTI:
Oh yeah, sure, that's a fun one to talk about. So, as you all know, Microsoft has what's called or you may not know this, but Microsoft has what's called power virtual agents. It's basically a platform that lets you create bots.

And so we are on the verge of, I call it, adding a new member to the marketing team, but we are creating a bot for internal use and we've given him a persona. His name will be “Mark Eting”. And so he will be going live real soon and we'll be rolling that out to the organization. But the role of this bot is for people to ask Mark, “Hey, where is the latest PowerPoint template or what is the latest? Oh, where can I get my business cards?” And so all the frequently asked marketing questions, this bot’s going to answer and of course if it can't find an answer then it's going to direct to one of the team leaders of the marketing team.

But, think about this, the way it's generating its response is that it's using, right, ChatGPT in the background in two ways. One, it is scanning our website, right, for the answer. So it's using that kind of indexing that ChatGPT uses to find the answer. But when I'm trying to give it a topic that it's outside of the realms of our website, it will use ChatGPT in the form of copilot to craft that topic. And so, it's just phenomenal what Microsoft is doing. And Sam, I know, speaking of artificial intelligence, I know that you have been using AI to come up with some pretty interesting graphical designs, and that's really what I want to talk about is, you know, what are you doing to pull up some of these images that are literally or they're work of arts, right. They really are. And they're unique.

SAM:
Perfect. Well, thank thanks for that intro and thanks for having me on as well to talk about such an interesting topic. And Santi, don't underplay your role in building the AI that you're building. I think it's really going to transform what we do in marketing.

GEORGE:
And that’s a good point actually, Sam. I mean more broadly and we'll see this as Sam jumps into this. Part of this is around being almost like a programmer, right? An operator of these tools and you need to be an expert operator to be able to get an actual usable outcome, and we're going to see it with what Sam did.

SANTI:
That's right.

GEORGE:
I'll make some commentary as he goes through some things, but in parallel I've been doing some of the using some of the same tools that Sam has in particular, Midjourney, but my output is not Sam's output, right? Sam's output is more refined, more consistent. You'll see it in a couple of things that he runs through.

SAM:
We will keep it quite simple today, but one of the interesting things I'm seeing in the market is full job descriptions for AI prompt engineers. And these aren't junior roles either. These are very senior important and key roles within companies, so while we're going to go through some basic prompts today, hopefully we can start to envisage how far they could go?

So as said, we use a number of tools for generative AI, some for copy, some for finessing copy, some for code, basic code. But the one that we'll run through today is for art generation. My predominant tool is Midjourney. There's a couple of really big players in the marketplace Midjourney and DALL-E. DALL-E is very good for realistic, photo realistic headshots, some more of the photos that you see on the internet that fall in the world where Midjourney’s really representing art.

Now a couple of interesting things, and if it's OK, I'd like to share my screen, yeah. So, I think a really key thing to note here is to access Midjourney you need to use Discord. Now Discord is a social platform for chats and forums. It was originally for the gaming community, but now it's more common for the artists in the tech industry as well. Because this is art-based, it's designed to use the community to teach each other, and that's the best thing about this. So, once you sign up for a Discord account, very simple to then add a Midjourney server. Once you sign up or Midjourney, it will invite you to the Midjourney server within Discord. Really, really self-explanatory - it took two minutes.

Before I get into the detail of what we do, I'd just like to show you the basic Midjourney server. Now what you'll see scrolling on my screen, it's not the art that I'm generating, it's the art that hundreds of thousands of people that are using this particular server are generating. So, I give you a really untested example of the general image gen server and you'll start to see the type of things and the type of prompts, which I will highlight here that people are using to generate their art. Now for newbies, which I was six months ago, this is vital to learning what works and what doesn't.

After a while it does get a bit tedious to see what everybody's doing, so it's very simple to set up your own server. And here I would like to show you the basics of how we generate our first images. OK, so “imagine” is the key prompt to get into art generation. We have a certain style of art that we would like to produce for our blogs, as one example. So, as I hover over them, you'll see that these are striking. These are bold. They've all got a certain theme split screen. The aim is around UCaaS and collaboration. We want to show that two people can be in, you know the same room or one person could be in two rooms, etc. So, we have built a certain level of prompt that creates this type of image for us.

But we need consistency and one of the problems with AI generation, it is quite hard to be consistent, especially when you have such a striking style. So, the way that we do that is by understanding exactly how we build our prompts. So just before the call, I created or wrote 3 untested prompts, from very very basic. Here, “split image of two offices one person in the middle”. Now you wait for it to do its thing. The greatest thing about watching it work is you start to understand how the AI tools render the art. It will always produce a grid of four choices. And you see it building, you see it layering the different styles.

SANTI:
Look at this thing.

SAM:
It's, it’s insane. And actually, it's one of those jobs that when you start, you can still be here 7 hours later because as you watch it build, you get fascinated with the detail that it produces. But as you can see really really basic prompt that kind of makes sense based on our artistic style that we're going for, but these are way –

SANTI:
And Sam, this is unique, right? Like there's – this it's, not like it's not – it didn't grab, it didn't Google an image and said here it is. This is – it's generated, right? I mean –

SAM:
Yeah, 100% unique. Well, obviously it's prompted by art that exists in the world, but the amount that it's looking at and changing it, it really is 100% unique. So, from a royalty perspective, if you have a paid Midjourney account which is only 10 bucks a month. 10 bucks a month for unlimited fast server and unlimited storage, unlimited generations, right? But once you have a paid account, actually, from a royalty perspective, you are allowed to use that, you own that. So, there is no, there's no clash out there.

So, here we are. We have some really interesting images. Some of them have bits of styling that we could potentially use, but it's too dull for us, et cetera.

So, then we look at the next level of prompt and you'll see as I type. I've literally just added a few keywords, so this time it's “the back of a young professional.” So, we're adding that layer of detail –

“looking away at a desk central in a split image of two very different offices.” Really, really basic still, but layered a couple of pieces in. Let's see what this generates. I said this wasn't tested, so hopefully we start producing some interesting layers.

GEORGE:
Getting closer, right?

SANTI:
Look at this.

SAM:
You can already see that it's getting closer. As you can see, up here highlighted here, it will tell you how quickly it's rendering and building.

SANTI:
Look at this.

SAM:
Now, if I if I just expand this. So, I would say if we look at these three grids or 4 grids, the one in the bottom right hand corner, it's quite close to where we're aiming, but the perspective is wrong, the colors are wrong. I think it's still a little bit grainy and not quite as realistic as we would like. So, then we have to get into the detail. So, I'm not going to be able to read this entire prompt out because it is too long. But there's a couple of –

SANTI:
I see. But this speaks to what you were talking about, George, right? And it's what we've always said. It's the input that that gives you the outputs. And that's where when we think about AI, it's about reinventing how we're going to use AI to be more effective and you have to kind of reinvent yourself to get ahead of game and be relevant in this world of AI, I mean this is it, right here.

GEORGE:
Absolutely. And the end benefit of this is speed clearly, right? This is much quicker than you could come up with a rendering as a graphics artist on your own and diversity of options. So, Sam and I have looked at various options over the last five or six months and the ability to come back within a couple of hours with eight or ten completely different variations of images, it's just impossible to do any kind of a normal graphic development environment. So, Sam, I realize this is like really complicated, right? What you've just typed in? Can you give us a couple of highlights? Not necessarily dissecting every piece here, but give us a couple of highlights of what does all this mean that we're seeing in the prompt?

SAM:
Of course. So, I think the first thing is the aspect ratio. So, as we're building for web and we need more of a landscape. So, there is certain prompts that you can put in that create a certain size of image, so you can see at the bottom. I want a 16 to 9 ratio, so that's really key and you'll see that in a second.

Here you can see the dash, dash, B5. That means that it's tapping into the version 5 the most up to date version of Midjourney which actually produces and sources slicker, slightly more focused images. Now the rest here, as George said, it's a long prompt so we can't go through it, but when we look at the image that didn't have the detail, we would want to add certain colors in. So, in the prompt I talk about the RGB of our three or four main palette colors. We want more white space. We want more creativity and abstract nature of the split between the two offices. So all of these details, and bearing in mind there's probably 30 to 40 individual elements put into this prompt, it's taken five to six months to get to that level where we can now repeat, they're very similar, but different based on these keywords, and it's all about where you put them in the prompt. Because the key is they need to look like the same family, but they need to be different. So, let's give that a test, and fingers crossed six months worth of work is paying off in a live audience.

GEORGE:
Let’s see what happens.

SANTI:
There's no post-production at this point.

GEORGE:
So, given the complexity of the prompt, will this take a little longer to generate or same amount of time?

SAM:
It's a really good question. It takes the same amount of time and when you have the 10 bucks per month license, it's generally very, very fast. As George mentioned earlier, I think with this and the writing tools that we're using, not only is it 10X style productivity, but it's, you know, 5X reduced our cost to produce this type of activity. Now whilst it's rendering for anyone that's just listening, you can already see and even though it's 60% through the process, you can already see that there is results here that are going to be interesting to us. You can see that the position of the main characters and the contrast and the colors and the palettes are going to be something that that we will –

SANTI:
Oh yeah.

SAM:
…want to generate. And hopefully if time’s allowing, I'm gonna then very quickly show once we pick an image how you can then constantly improve and repeat that image also very quickly.

SANTI:
Look at this. This is amazing.

GEORGE:
And that was the challenge early on, Sam, right? It was a little more difficult than the first couple of tries at this going months, you know, back in the 2022 to get the repeatability that we wanted.

SAM:
Exactly that. So, I would say here on our grid of four, the top right is probably the best and the most the most like our current feel. So, you do two things we upscale that particular image so then it takes that image. And, oh I've upscaled the wrong one, sorry upscale 4. So, I'm going to upscale 2. It takes that image, re-renders it and adds another layer of detail, so then it's web ready. So, it's a very, very high quality file. But then if you are still not quite happy with it. Because no matter how advanced these tools are sometimes there's still some oddities. There's still some things that you aren't very happy with. So, you can literally just make infinite amount of variations of that one particular image. And sometimes it changes the entire, maybe computer that's sat on this gentlemen's desk, or sometimes it changes the -- look at this perfect example – so, looks like one of the computers has been completely removed or turned into a laptop.

GEORGE:
And I know early on, we had challenges. People with six fingers and you know, with the leg of a chair missing or in the wrong place, but it seems like we've seen far less of that in the most recent iterations.

SAM:
Yeah, it's a really good point to bring up. So, there is still some that aren't quite perfect. So, fingers and legs, they're very strong now. Sometimes faces have a, you know, a very, very slight oddity. What's really important here, and I think this is exactly the same for all of the copy we wrote and all of the code that we write, whether it's basic code or really advanced, using AI, we are getting closer and closer by the day to the finished product. But, here is still a really, really key layer of human touch and finesse that is required at every single stage. So, we would then take our favorite image from here and we would go and add our overlays and our finessing and potentially some Photoshop if necessary. And it's the same with the code we have to go and you have to go and clean up the code. But, there we are. This is one of the five or six tools that we use and this is how much time we can save. Getting to the point –

SANTI:
It's amazing.

SAM:
…that we would never have got to before AI, because I think this has opened up so many doors for us to be a bit braver with the style of content that we produce.

SANTI:
Wow. Well, let me tell you, Sam, this, I love this stuff, but this is mind blowing and to see it, like, literally developed the image before your eyes, that's so telling. And I'm blown away by this, and as usual, right, we could talk about this forever, but we can't. So, we have to bring this podcast to an end. Sam, thank you so much for joining us and for taking time to show us this really, really cool tool.

GEORGE:
Yeah. Thanks Sam. This was this was amazing.

SANTI:
That was awesome, that was awesome.

SAM:
Anytime, Gents. I enjoyed it. Thank you.

SANTI:
So, to make sure that you are alerted on any upcoming episodes, please make sure to subscribe to this podcast on your favorite platform. And if you want the show notes for this episode, you can find those at www.fusionconnect.com/techUNMUTED. Until next time. Stay connected.

CLOSING VOICEOVER:
Visit www.fusionconnect.com/techUNMUTED for show notes and more episodes. Thanks for listening.

Episode Credits:

If you want to give shout outs to specific people who helped with the episode.
Fact-checking by: Joe Jimmy Jim
Additional Video Editing by: Some Famous Person
Produced by: Fusion Connect