Jamey Gannon: All right. Hi everyone. Um, today I'm going to show you a very, very simplified system for making basically any AI product photography that you could possibly need for your brand.
> [VISUAL: The screen transitions to a presentation slide. A small video of Jamey Gannon is in the top left corner. The slide is titled "AI Product Photography for E-Commerce Teams" with the text "With Jamey Gannon" and four example product photos below the title.]
Uh, we only have 30 minutes today, so I'm going to run through sort of like my favorites and the most core things I think you may need.
> [VISUAL: Slide with four images of a black jar labeled "PORTAL". The first is a standard product shot. The second is a studio shot with a blue gradient background. The third shows two jars floating. The fourth shows a man on a couch reading a book with the jar on the table.]
Um, so we're going to go over starting from a label design or just like a basic iPhone photo. We're going to do creating high quality studio photography, creating editorial and stylized photography, and then also creating AI UGC and inserting your product into it.
> [VISUAL: Slide with a photo of Jamey Gannon winking. Text on the right: "Jamey Gannon @jameygannon (aka techbimbo)".]
But before we dive in, I got to want to share a little bit about who I am. So, my name is Jamie. If you follow me on X, I'm kind of known there as Tech Bimbo.
> [VISUAL: Slide with a black background titled "I have like 4 jobs". It shows a montage of images and text labels describing her roles: "Freelance Brand Designer and (AI) Creative Director", "Head of Brand @ Portal (supplement company)", "Teaching creatives how to be AI Creative Directors", "Content Creator", "Leading social @ Pomelli by Google". The montage animates quickly, showing various branding projects, websites, and social media content.]
Um, and I don't know about you guys, but I find it harder and harder to introduce myself like every week. Um, job titles are really starting to blur. But anyway, if you've never come across my content before, I am a brand designer and creative director who uses AI every single day and I basically make a ton of content talking about it as well. So, my background is primarily in early stage companies across tech and D2C, but I now sort of split my time between those actual branding services, like this GIF that you see on the screen, and then helping companies and individuals bring generative AI into their creative workflows. So, doing workshops like this, um, talks, and then I also have a course called The AI Creative Director, which is actually starting its next cohort in July. Um, and then I have some side jobs if you can call them that. Um, I actually have a supplement company with my boyfriend called Portal. It's a sleep supplement. And then I also, uh, lead social at a Google Labs experiment called Pomelli, which is another really cool AI tool, um, for business and e-com stuff. Um, but yeah, basically I've gotten to work with loads and loads of people and teams across like every single intersection of design, business, tech, AI. Um, and I've actually even worked with like some of the speakers, um, today as well on this very stuff for their brands. Um, yeah, so that's who I am. Uh, let's dive into it.
Slide with a black background showing a 3-step process. "01: Label Design" shows a flat, unwrapped product label for "PORTAL". "02: Label Design" shows the same label but cropped to just the front-facing portion. "03: Mock-up creation" shows an image of a competitor's product ("ARMRA"), then an arrow to a blank black jar, then an arrow to the final mockup with the "PORTAL" label applied.]
So, uh, I'm going to be showing you guys a couple techniques, basically is how I think of it. And the first technique I want to show you guys is creating a product shot from just a label. And I love this use case for a few reasons. Number one, it's great if you're just trying to iterate on a new design or like a new SKU. And two, with this technique, you can actually start making images of your product before it's even in production. And I think this opens doors for so many things like paid ad testing, doing pre-sales, and also just moving faster in this market, which I know is probably top of mind for a lot of you.
So, to break this down, firstly, this software that you're seeing, anytime you're seeing like a dark mode software, um, in this talk today, it's going to be Flora. Uh, and Flora is just a multimodal tool. They're very popular for their node system. Um, but this workflow is like really tool agnostic. Um, there's so many incredible AI tools to use these days. Uh, I use almost all of them. Um, you have like Weavvy from Figma, you have Pixfield, Krea, Magnific. Uh, for the product design or for the product photography use case that we're talking about today, the models are primarily going to be using our GPT Image 2 or Nano Banana Pro. So, any tool where you can access those models is going to be totally fine for these techniques.
Um, but yeah, the first thing you're going to want to do is get your label design. So this is just a PNG export of, uh, my Illustrator file, like my actual packaging design. And then I just place it in the top here. And then the next thing you're going to want you're going to want to do is crop it and you just want to show the part that's actually going to be in the image. Um, so make sure you take out like your supplement facts, your story facts, or whatever else you have on your packaging that you don't want on there. And then next you're going to need your actual mockup. So in this example here, I actually didn't have a picture of the plain black jar for whatever reason. I couldn't find it from my supplier, I couldn't find it online.
> [VISUAL: Slide showing the process of removing a label. On the left is the "ARMRA" product. An arrow points to the right, where a plain black jar is shown. Below the plain jar is a text prompt: "Remove all design from this canister, make it a plain matte black canister."]
Uh, but I remembered that Armra has an extremely similar, probably the exact same packaging that we do. Um, so I just took Armra's, um, and I did a very simple prompt just telling the model to like remove all the text. Um, so what I say here, I said, remove all the text from this canister and make it a plain black canister. And you can see, it just did that.
Slide showing the final step. The cropped "Supplement Label" and the "Matte Black Canister" are shown as inputs pointing to the final "PORTAL" product image. A long text prompt is displayed on the right.]
Then from here, I just created a new image block and I connected that mockup that I made, and then I connected that cropped label design and then I gave it this prompt. Um, you guys can read this, I won't read it to you. But there's a couple things that I want to call out on this prompt here. So you can see it's quite detailed in terms of instructions. And this is something that you will learn from trial and error with your exact specs for your product. So, I've tried to make this image many times over the last few years with all these different AI tools. And I noticed that, you know, there's a few things it constantly messes up like the the texture of the label, the exact color of the label, um, the color and material of the text. Like sometimes AI will just make it not the way it looks in person. Um, but over time, I now kind of have this like master prompt that I use, uh, whenever I'm doing imagery of this particular product. And over time for your product as well, especially if you have like really unique form factors, um, or a super custom packaging, like if you're a perfume or something, you'll need to figure out like what does the model need to know every single time, because otherwise it's just going to like default to kind of like what it knows.
> [VISUAL: A close-up of the final generated product image of the "PORTAL" jar.]
Um, and often times it won't get this right. So that's one important note. Um, and then here we have our final result. Um, I wish I should have put a jar of portal here, but this is exactly what it looks like. Um, it's matte, it has the metallic text. Um, and yeah, I think this was a pretty good result. Now, what we just did may seem simple, but we actually made some intelligent and considered decisions when running through this process and I want to break them down for us.
Slide titled "Why Nano Banana Pro/GPT 2?". It's split into two sections. "Diffusion" on the left shows three artistic, abstract images. "Multimodal" on the right shows three realistic photos of people and a product.]
So first, why are we using Nano Banana Pro and or GPT2? And why are they so uniquely good at product photography and realistic imagery like this? Um, so both of these models are actually multimodal models, which means they can actually think and understand what you're asking for. And this is different from diffusion models like Midjourney or Flux that are more optimized for aesthetics.
> [VISUAL: Slide titled "Starting Conditions / Inputs". It has three columns: "Model Selection" (with examples for Diffusion and Multimodal), "Reference Material" (with examples for Subject, Scene, and Style), and "Prompt" (with examples for Vibes and Specific).]
And basically, if you're using a multimodal model, or if you aren't using a multimodal model, you should just not expect it to keep your subject or your text, for example, exactly the same. So whenever you see something that involves extreme precision or instructions, like mockups or text, or even these images I made of myself here, uh, as of right now, it's going to be a multimodal model like Nano Banana or GPT Image 2. Um, now I consider multimodal models one of many starting conditions you have for your generations. And the first starting condition is model selection, which we just went over. The next is your reference material. So models and tools handle this differently. Um, for most of them, it's image uploads. Sometimes it's mood boards or codes if you're using Midjourney, but they're used to do three things: to establish your subject, like an object or a character, to establish your scene or establish your style, like illustrative versus photorealistic.
The slide from 1:56 reappears, showing the 3-step process for creating the product mockup.]
And then lastly, the one that we all know is your prompt. And depending on your other starting conditions and sort of what your goal is, um, you can do more vibey short prompts or longer, more instructional prompts. Um, next, you might be wondering, Jamie, if AI is so good, like everyone has told us, I don't know about today, but like I feel like the narrative is like AI is incredible and it's taking over everything. Um, you know, why did we have to work in so many steps like this? Why couldn't we just like one shot it or just like do one prompt? Um, well, the truth is, AI is really not that good yet, at least not for everything and not for complex tasks like this. Um, so this I want to kind of talk about some of my principles I actually teach in my course. I have five main ones.
> [VISUAL: Slide with the number "04" in the top left. The main text is "Work in steps". To the right is a photo of a woman in a white dress running up stairs, her blonde hair flying.]
Um, and one of them is that you should always work in steps. Now, if you've ever been in like a coaching call with me, or been in like a session where I'm like reviewing prompts or helping people work through why their prompts aren't working, you'll probably hear me say this 10 times. Um, but basically when you are dealing with complex tasks, it's always going to be easier to tackle the problem in steps, whether or, you know, when you ask the model or even yourself to do many things at once, your prompt adherence is going to be lower and you're just going to have more variables to work with. But when you work in steps, you're able to nail your elements one at a time and it'll just make your life and your generations a lot easier. Now, you can definitely break the rule with this depending on the model and the task and like how particular you are about your prompting, but I would say as a best practice, most of the time you should work in steps.
Slide showing a failed one-step attempt. The "Sleep Supplement Label" and the "Nutritional Supplement" (the ARMRA product) are shown as inputs pointing to a poorly generated "PORTAL" product image where the label is distorted and incorrect.]
Um, so for example, I want to show you what it looks like if we do the exact same task, but just jump straight into it. Um, so I'm using the same image, the same label, the same model family, and practically the same prompt. Um, the only thing I did was say remove the text on the first image. Um, and you can see how much lower the prompt adherence is. So it's inventing details, it's missing colors, and like overall even the image just doesn't look as good. Like the aesthetic of the actual jar is kind of bad to me. Um, so, yeah.
> [VISUAL: Slide showing a process. An iPhone photo of a "PORTAL" jar on a table is the input. The output is a high-quality studio shot of the same jar. A long, detailed prompt is shown on the right.]
Okay. I want to show you guys a very similar task and output, but we're going to approach it in a different way, and I think that this one will be like much more common as well for you guys. Um, so we're still making a studio quality image with a white background, but this time we're going to start from an iPhone photo. So, as you can see here, this is my only input and it's just this image and this prompt in Nano Banana Pro. And this is the result we got.
Slide titled "SWORD" with an acronym breakdown: Subject, Where, Orientation, Rendering, Directives.]
So, it's pretty clean, I would say. And let's break down this prompt. So, when I'm prompting for specific things, like the last thing we just did, I am unconsciously and sometimes consciously doing what I call the sword framework. Um, so S, which is the subject, which is who or what is in the image. We have W, the where, so the setting, the scene, the background, the placement. We have O, orientation. So it's like pose, framing, product position, body direction. We have rendering, which, um, is kind of like camera style, lighting, color, texture, um, is it realistic? Is it illustrative? And then D is for directives. So these are like the rules, the constraints, common sense checks. Um, you know, when you're saying like, don't change anything, those are the sort of instructions that you have to add.
> [VISUAL: The prompt from the previous slide is shown again, with different parts highlighted as the speaker talks.]
And I love this framework because it's extremely legible for even the least like creatively minded people on your team, like someone that doesn't necessarily think of themselves as an image maker or creative director. Um, and it's a really good reminder of what details AI needs to be given. Uh, you can't just like lazy prompt everything, especially when you're trying to be as specific as like product photography. Um, so let's like break down this prompt a little bit more. Um, so here we have our subject, which is matte black supplement jar with metallic text. Next, we have our where, which is a white photo studio environment. And then the orientation here is kind of sneaky. So, you may think saying product photography shot is kind of obvious, but it's actually sort of a shortcut that we're giving the model. So, the model has access to the internet and, you know, training, it knows what product photography shots are and it knows that they're typically very high quality, very photorealistic, very straight on, not typically obscured or abstract. So, instead of saying like centered in the frame or other attributes that you want in a product photography shot, saying product photography shot just does that for us. Um, our rendering aspect are these details here. So we're saying ultra high quality, strong directional strong directional natural light, and the camera type and the style. And then lastly, we have our direction. Um, so this paragraph here is actually what I call one of my prompt boosters that I throw on any generation where I'm dealing with text or logo or product consistency. And this basically just says like, don't change anything. Um, and for whatever reason, this helps my generations be like 50% more accurate. Um, and I think as you start to develop your own like custom workflows and like I said, finding unique things that your brand or style or product needs, having these just like little prompt boosters ready to go all the time, um, can be really helpful.
Slide showing a process. On the left is a photo of a Byredo perfume bottle and a photo of a "PORTAL" jar with dramatic lighting. Arrows point to two generated images on the right, both showing the "PORTAL" jar in the style of the Byredo photo.]
Okay. Next technique I want to show you guys is how you can achieve really aesthetic stylized editorial images like this, but with your own products. So, all you're going to need to do is find an image with the style you love. It doesn't actually have to be a product photography image. This is just like an easy example. Um, it could be literally any photo that you want. Um, and then you're basically just going to tell AI to create an image of the product in the first photo in the same style as the second reference photo. So, make sure you hook these up or upload them in whatever order you write the prompt in. Um, and then of course, this big paragraph here is that prompt booster again. Um, so a few things I want to note. Uh, we're using one of the iterations of that really clean white background studio shot. So, in this case, mine has some shadows to add just a little bit of like flair. Um, and you can tell that we're kind of going back to that work in steps principle. So, instead of having to tell the model to, you know, remove the background and make it a high quality image, like take it from iPhone to studio to then the aesthetic studio, we're already we're just skipping a few of those steps by giving it like a plain background studio shot. Um, and then for this example on the screen, I tried it in both Nano Banana Pro and GPT Image 2. You can see that the results are both great. Like technically, I think the GPT Image 2 one on the bottom was a little bit more faithful to the style, but for better or worse, they're both good results. Um, and I encourage you, you know, if you are going through a process and maybe something's not working or you just want to see like what's going to give you a result you like better, feel free to like try multiple models and multiple prompts.
> [VISUAL: A new image appears, showing two cylindrical containers, one brown and one green, with the brand name "DALGONA" on them.]
Um, I have a typo here. Oh, a composition. A composition like this would be really great for, uh, you know, if you to extend it in like 9 by 16 for ads, you could put a bunch of text on it. Um, or even like a hero shot for like a landing page. Um, and I really love how easy it is to get this like really like physical type of imagery that previously would have needed like a serious understanding of like spline or another 3D software. And you can just get this done in like a couple seconds really.
Slide showing a grid of various product photos in different styles: a perfume bottle, a soda can, a sunscreen tube, a dropper bottle, and a can held by a person.]
So, you can basically rinse and repeat these techniques with almost any reference or idea or object. Um, you know, we're just see here like stealing the style. Um, you can even like have it be floating again, it could be sat down. You could even take like this uh, this dropper here, maybe even swap it for this one. Um, but at the end of the day, all you're doing is uploading images and then telling it how to use those images.
> [VISUAL: Slide showing four images in the style of Instagram stories. They are selfies of different people holding or interacting with products.]
Okay. Let's move on to UGC. So, for this section, we're going to be using a combination of Pixed Soul and then again, either Nano Banana Pro or GPT Image 2. So, the reason I love Pixed in particular for this task is two reasons. Um, number one, their model Pixed Soul. It's just extremely realistic. I don't know how they trained it, uh, or if it's like a layer on top of another model, but it's really, really good. And then two, they also have tons of templates and essentially shortcuts that are particularly great for e-com use cases, uh, when it comes to like content and consistent characters, um, and just like social-esque stuff.
A screen recording of the Pika Labs interface. The user navigates through different models and presets.]
Um, and we can't like get into everything today, but I do want to encourage you to like go and play on that platform because it's kind of bonkers. Um, so Pixed is a really big platform. Um, so I'll show you exactly what you need to do. Um, you're going to go to this like farthest top left menu here. You're going to hit image and then generate image or create image. And then you're going to want to go down to the bottom and make sure that you select Pixed Soul as the model, not Pixed Soul 2.0, it's different. So just regular Pixed Soul. Um, and then you can actually adjust these like style presets. Um, they have tons of different ones. Uh, for UGC, I'm typically using like 0.5 selfie or iPhone or general. Um, but they also have like, you know, realistic, they have flight mode, which is like airport, like hotel mode, they have like Japandi, uh, like Tik Tok aesthetics basically, which depending on your audience can be really helpful shortcuts.
> [VISUAL: Slide showing a process. On the left is a photo of a man reading a book on a couch with a black candle on the table. An arrow points to the right, where the candle has been replaced by the "PORTAL" product. A text prompt is shown above the final image.]
And then another great thing about Pixed is they have this incredible prompt rewriter in the back end, which I typically don't love, but for this use case, I think it's really, really helpful. Um, so for example, to get this image, my prompt was something really simple. A man reading a book on his couch in an apartment. In front of him is a matte black candle on a coffee table. And then Pixed will then also using that style selector, rewrite the prompt in the back end to be much more detailed and have things that feel super natural and things that you actually want. Um, so it describes the texture of the candle, it describes the quality of the lighting in the room, it describes his outfit, which we never mentioned, um, all this stuff.
Slide showing a process. On the left is a photo of a woman in pajamas on a chaise lounge in a high-rise apartment at night. An arrow points to the right, where the prompt rewriter has generated a very detailed descriptive paragraph based on a simple initial prompt.]
And a way that you can make this really work for you is by having choice words in your prompt, like luxury, for example. Um, so with this image, again, we did a very simple prompt, but because we used the word luxury, we saved a lot of time in actually having to describe what luxury might look like. So, again, AI understands the world, it understands what luxury is. It knows it's going to be a certain type of woman, a certain style of clothes, a certain style of interior design. We can see that she's obviously in a high rise in a city with lots of windows and nice drapes. And this sort of prompting actually works for every model, not just Pixed Soul.
> [VISUAL: Slide showing three images in the style of Instagram stories: a man on a treadmill, a man sleeping, and a woman holding a product.]
Um, so luxury is like one that's really easy to use. And just getting specific with your prompts in that way can make you like a good lazy prompter, I'll say. Um, and yeah, Pixed Soul is great at pretty much everything you can see here. Um, a lot of it looks like terrifyingly realistic. Um, but you'll notice if you look closely and you're not just like viewing these as like super quick ads, there's definitely a bit of slop in them. And then of course, our product is not inserted yet.
Slide showing a process. On the left are two images: the "PORTAL" product and the photo of the man reading with the candle. Arrows point to the final image on the right, where the candle is replaced by the "PORTAL" product. A text prompt is shown.]
So, I want to show you this as an example. And if you'll notice, I actually did a bit of pre-work in my original prompt. So, you can see here, I purposely asked for this image to have a matte black candle. And that's because I found over time, um, when prompting to make images of the portal jar, if I said like supplement jar, it would sometimes give me like something very small, or it would give me like a protein jar or like odd shapes. And then it was just like another step for AI to work through, um, when I actually went to go do the insertion. Um, but one day when I was looking for inspiration for product photography, I realized like, oh, candles are like so similar to like the actual visual weight of Portal. So now I started doing this as a shorthand. And then when I go and do the replacement, um, it like just works a lot better.
> [VISUAL: Slide showing a process. On the left is an iPhone photo of the "PORTAL" jar. The output on the right is a high-quality studio shot. The same jar. A long, detailed prompt is shown.]
Um, another thing I have in this prompt here, um, the the prompt is like pretty self-explanatory. Um, but I also have like adjust the angle and the lighting to make it fit seamlessly with the scene. Um, sometimes the models can be a little bit dumb. If you say like insert this product into this image, it might just like superimpose it like a sticker. Um, but if you tell it to like actually, you know, make it look realistic, it can tend to do it better. And that's another benefit of having an object that looks similar enough to your product to begin with. It's not having to think so hard like what the lighting is going to be across the shadow and the shadow it's going to cast and things like that.
Slide showing a process. On the left is the photo of the man reading with the "PORTAL" jar. An arrow points to the right, where the book cover has been replaced with the cover of "Can't Hurt Me" by David Goggins. An image of the book cover is shown as an input.]
Um, if you have like a supplement bottle that's like just like a plain white bottle, you can probably just do plain white bottle, but again, experiment and you may need to get creative, um, to get like the best results in this case. Um, and then when it comes to fixing slop, you can do like obvious things, like just telling it to like fix the shirt or change the color of something. Um, but you can also fix slop in a way that like really gets at your target audience, especially if you're making this UGC for like, uh, more evergreen content on your website. Um, so because he's reading a book in this image, I thought, hey, why not make the book something my target audience might like. Um, so I just like grabbed the cover of David David Goggins' book, um, on Google and just told it to replace the book in the photo with the one in the reference image. And then he's reading the David Goggins book.
> [VISUAL: A slide with a black background showing a grid of images. Text overlays the images, making them hard to read. The speaker seems to be describing a different slide.]
Um, oh no, there's text covering this. Um, and you can basically use the same flow again, whether it's like clothes or people or locations. Um, but again, while I'm doing this, I'm doing this very smart. So, my poor text. Let me see if I can remove this.
The slide from 20:05 is still on screen, with the text overlay issue. The speaker continues to describe a different process.]
Uh, anyway, it's a, it's a guy in like a shirt with a Nike logo, but the Nike logo was like slopped up. Um, so I just told it to, uh, replace the shirt that he's wearing with this one, and then it looks more realistic.
> [VISUAL: The slide from 00:17 reappears, showing the four "PORTAL" product photos.]
Um, so after just about 23 minutes, we have four totally new from scratch product photos.
Jamey Gannon: And realistically, you could make like dozens of these in the same amount of time that we've been talking. Um, especially when you get into like certain automations and different tools and utilizing like the nodes and workflows. There's a lot of crazy stuff out there. Um, and more is definitely coming down the pike. But for this workshop, I wanted to focus on what's actually happening in the back ends of the models and get you like the basics. So as you start to use these more templated tools, um, you know what's going on and you know how to fix it. And yeah. So that's the workshop.
> [VISUAL: Slide with a large QR code on the left and an image of a smartphone displaying a social media profile for "Jamey Gannon" on the right. Text on the slide: "Resources, Tools, Discounts, Social" and "www.brand-sprints.com/links". Music starts playing.]
Um, I guess I have five minutes to do a Q&A. Why is the music playing? Hold on.
The music stops. The QR code slide remains on screen. A second video feed of a man, "Evan", appears in the bottom left.]
Uh, you can find me pretty much anywhere at Jamey Gannon. Um, if you also head to like my, my link tree thing, uh, brand-sprints.com/links, you can find loads and loads of resources, whether it's my course, templates, um, I have like 25% off discounts to Flora and pretty much all the AI tools. Um, and of course, you can keep up with me on social or maybe we can even work together. Um, yeah, I guess I have five minutes to do a Q&A.
Evan: Amazing, Jamie.
Jamey Gannon: Um, I'm just reading some of these questions out. How do you deal with the negative feedback from social when they figure out it's AI? As a creator, I don't care. As a brand owner, that's something that you need to to think about and have a comm strategy for.
Evan: Yeah.
Jamey Gannon: Um, the name of the platform is Flora. That was the one I was using for the most part. And then the AI UGC one is Pixed Soul. Um, where would I recommend to start for beginners using doing AI image generation? Um, I'd follow me on X. Um, and then just start doing stuff. Like try and find a simple task, um, maybe like one of the examples I did today. Um, and just start small. I think AI is just like any other tool. So, you know, you didn't learn Google Ads or Illustrator, um, or CapCut in a day. You know, it took you time. Uh, you made a short video first, one logo, uh, you know, you did your first campaign. Um, so yeah, a couple hours a week, just start getting into it and then slowly you'll start to see more use cases for it.
A question from "Jose Garcia" appears in a purple banner at the bottom of the screen: "Some of the biggest issues when generating AI assets at scale is that, when the product is a little bit tricky, the agent will not do a very good job at maintaining the product consistency over time. Is there a way to prevent that?"]
Evan: Jamie, I'll pull one final one up from and put it on the screen here. From Jose. So it seems like he's he's using it quite a bit already, but some of the biggest issues when generating AI assets at scale is when the product's a little bit tricky, the agent will not do a very good job at maintaining the product consistency over time. Is there a way to prevent it?
Jamey Gannon: Yeah, I think there's specific tools that are starting to get better at this where you basically train a mini model on your product. Um, but as good as AI is, like it definitely isn't magic. Um, there was like one example I did, I did like a three-hour version of this workshop and someone, one of the students, of course, had a like optical illusion checker print toilet paper holder. And I was like, dude, I don't know. Um, and then we kind of we kind of worked through it like telling the model it's a toilet paper holder, um, creating like a shot list, like there's something you can do where you have, uh, you ask Nano Banana or you take a photo of it, um, in like a bunch of different angles and then every single time you like are using that as a reference image, it has like all of those. Um, I think it'll get better over time for even more complex things. But if like, like even that image that she gave me of the checkerboard toilet paper holder, like I wasn't even totally sure what it is. So I'll say like, like yes, AI is very intelligent, but it's not going to like read your mind. Um, so you might just need to like have a very long prompt, have more reference images. Um, also, I Photoshop so much. Like I, um, one of another one of my principles is use the right tool for the right job. And we're still at a point where like sometimes to like edit a logo or add a fingernail or something that gets lost, especially really tiny stuff, you just got to have to Photoshop it. Um, now if you're if you want to try and generate hundreds of images in a day, totally different thing, but if you're trying to make like really high quality images, um, you know, going into Photoshop is there's nothing wrong with that. It doesn't mean that you're doing it wrong. That's just how things work right now.
Evan: Huge. And Jamie, any final words you want to leave with the audience today?
Jamey Gannon: Uh, persevere. This stuff is really hard. People want you to believe that it's not hard. People that sell courses like me, they want you to think it's very easy to keep generating stuff. The models want you to think it's very easy to keep generating stuff, but it is a tool and a skill just like any other. And it will take you time.
Evan: Amazing.
Jamey Gannon: And you'll get better at it. And then you'll be able to do it in 20 minutes.
Evan: Amazing. Thank you so much, Jamie. That was an incredible presentation. I think everyone here got a ton of value out of it. Um, and yeah, we'll see you on the next one.
Jamey Gannon: Yeah.
Evan: Thanks, Jamie.
Jamey Gannon: Bye.
Evan: Bye.
