Scaling DTC Ad Creative with JSON Prompts and Veo 3

1 Aug 2025 · 51 min · 23 chapters

Ask about this episode

Ask anything about it. ChatGPT or Claude reads this page and answers with the times it was said.

Connect VO and ask about every podcast you hear, including the moments you saved. Add to ChatGPT · Add to Claude

In short

How to use Google’s Veo 3 (VO3) with JSON prompts to scale DTC ad creative—especially paid social video ads—by leaning into satire/comedy, generating extreme-but-brand-aligned concepts, and rapidly iterating hook variations. Workflow covers brainstorming in ChatGPT, converting ideas into VO3-ready JSON (camera, shot type, movement, location/time, style, scene/subject, dialogue/emotional tone, and constraints like “no subtitles/text/logos/music”), and operational tips (8-second scene blocks, VO3 fast vs quality, trimming prompts, character consistency via “anchor”/celebrity references or frame-to-video).

Guests

Lucas Hamby (AI/creative “extraordinaire”; described as “at one with the robots”). Host: DTC Diaries creator (unnamed).

Key claims

realism isn’t always the goal; “lean into the ridiculous” works; VO3 can 10x visuals by reusing account learnings; hook testing is the fastest starting point; VO3 fast is preferred; character consistency is crucial.

Notable examples

“Space Goods” 80s neon supplement ad with POV “powder on a spoon” hook variations (e.g., rollercoaster/roller coaster spoon, split-screen); Motion-style satirical “podcast ads” with babies (fourth-wall comedy; voice sync limits); Too Short for Modelling’s Liquid Death commercial; story-driven Space Goods montage; hook variations for creatine/menopause using whiteboard-style concepts.

Written by AI. May contain mistakes. Listen to the episode to check what was said.

Chapters

Tap a time to open that second in VO

Exploring VO3 and Its Potential

0:40 to 3:19

Discussion on VO3's features and how to effectively prompt with it.

“Make sure you're subscribed if you're not.”

Creative Philosophy in AI Ads

3:19 to 6:44

Insights on creative approaches and the importance of mindset in AI-generated ads.

Examples of Innovative Ad Formats

6:44 to 8:40

Examples of unconventional ads using VO3, including comedic and satirical skits.

“babies on a podcast so let's give this a quick watch um ai queer strategies straight fire a hundred percent what's your favorite task to run in motion i like generate winning ugc scripts oh Oh, that slaps.”

Story-Driven Montages in Ads

8:40 to 12:16

How to create engaging story-driven montages while using humor and absurdity.

“And I don't know if you saw this, I think I sent it to you, but it's these guys called Too Short for Modelling.”

Leveraging Humor for Engagement

13:40 to 14:02

Final thoughts on using humor in ads and connecting brand identity with creative content.

Layering Dynamics in Creative Marketing

14:02 to 14:31

Learn how to use different characters and comedy in marketing to maintain audience engagement.

Aligning Brand Identity with Creative Elements

14:31 to 15:12

Discover the importance of aligning creative elements with brand identity for effective marketing.

Testing Hook Variations for Effective Ads

15:12 to 16:00

Explore how to create and test multiple hook variations to optimize ad performance.

“Or if we've got a creative that is showing good promise, but maybe the hook isn't, the thumb stop ratio is sitting below 30%.”

Generating Strong Hook Variations

16:00 to 18:11

Understand the process of generating effective hook variations based on audience feedback.

“My mum said to me the other day that Create Team was her biggest hack during menopause.”

Creating a Workflow for High-Quality Output

18:11 to 19:05

Learn how to establish a workflow for generating consistent and high-quality ad outputs.

Show all 23 chapters

Using ChatGPT for Creative Brainstorming

19:05 to 21:11

Discover how to leverage ChatGPT for brainstorming ad hook ideas effectively.

Structuring Prompts for AI in Marketing

21:11 to 24:32

Learn how to structure prompts effectively to avoid common pitfalls in AI-generated content.

“this is where it could say GoPro you know it could say like a really high-end camera it could even be the iPhone.”

Best Practices for Video Content Creation

24:32 to 28:00

Understand the best practices for creating engaging video content using AI tools.

Using JSON Prompts for VO3

28:00 to 29:10

Learn how to effectively create prompts for VO3 using JSON format.

Tips for Effective Prompting

29:10 to 33:10

Explore strategies for effective prompt creation and scene consistency in VO3.

Challenges with Product Integration

33:10 to 34:20

Discuss the challenges of integrating products into AI-generated content.

Optimizing Visual and Brand Representation

34:20 to 38:10

Learn techniques for brand representation and visual storytelling using AI.

Navigating Copyright and Creative Boundaries

38:10 to 41:40

Understand copyright issues and creative boundaries when using AI-generated content.

Future of AI in Content Creation

41:40 to 42:04

Speculate on the future possibilities of AI in personalized content creation.

Navigating Copyright Risks in Ad Creation

42:04 to 43:32

Learn about the importance of copyright considerations when using AI in ad creation.

The Future of Personalized Content

43:32 to 45:28

Discover how algorithms could redefine personalized media experiences.

“And I think I went into VO3 a little bit lackluster because I went to film school and I'm a creative person.”

Legislation and AI in Advertising

45:28 to 47:53

Explore potential legislative challenges and ethical considerations for AI-generated content.

“Because otherwise it's like, there could be infinite amounts of just lies on the internet of like fake people selling products they've never consumed to an audience that doesn't know they're not real.”

Impact of Technology on Marketing

47:53 to 49:22

Understand how new technologies like VO3 are transforming marketing strategies.

Hear the part that matters, and keep it.Open this episode in VO. Double tap your headphones to save a moment as you listen.
Get VO free

Transcript

Automatic transcript. May contain errors.

0:00So we're going to go through some key things to know when using VO3 and some kind of core learnings there, how to best prompt for it as well.

0:15Welcome back to another episode of DTC Diaries. Back with another solo duo episode. They go down well. They go down very well. Yeah, I think we need to do these more. We've got loads coming up in terms of just valuable stuff that we're doing. I think a lot of what we are learning internally and sharing with the team, we are going to be sharing externally on DTC Diaries. So stay tuned. Make sure you're subscribed if you're not. Yeah. And I think if you're listening to this episode, head over to YouTube for this. It's going to be a visual show and tell. I'm in the passenger seat for this one. We've got AI extraordinaire, Lucas Hamby.

0:53I don't know about that. um he is it he has he is at one with the robots at this point um we're going to be going through v v o three uh gemini um we did have a bit of a loose idea of trying to every time we cut back to lucas trying to turn him into a different animal or character to really make this as interactive as possible and then he told me how long that would take to prompt so not quite that engaging but hopefully value-packed nonetheless anyway yeah so today we're going deep into VO3. Obviously, Google released their new video gen model, which is a complete game changer really in many ways, especially indeed seeing how it can be used for lots of different things that

1:40Loukas:we're going to get into. So agenda today, we're going to go through use cases firstly for how we can utilize VO3 as part of paid ads. This is stuff we've been diving deep into over the last month um key things to know i've been like probably spent near 24 plus hours over the last um well way more than that actually over the last couple of weeks just just really going headfirst into it i blocked out a lot of my calendar over the last few weeks just to you know really absorb it as much as i can so i think there's some good learnings that i've taken that can apply i think one thing i will say as well is make sure you're subscribed because we're going to be giving away um a load of amazing prompts that i've been really stress testing over the last couple of weeks um for things like hook variations um you know concept ideation with uh chat gpt before we bring it into vo3 um and and a few others so so yeah make sure you you get over to youtube and subscribe drop a comment robot and we'll send you one of those prompt guides over um it's funny because every episode we never decide what the comment is we just like one of us just comes up with it on the spot feels um apt for this episode um yeah so it's so we're going to go through some key things to know when using vo3 um and some some kind of core learnings there how to best prompt for it as Well, creative philosophy.

3:12So just things to keep in mind, you know, when we're actually using it, because I think it isn't just about the output.

3:18Loukas:It's about the script and how we use that script, how we lean into comedy, jaw dropping elements, obscurity, emotional resonance, things like that. yeah i think i think one real fundamental like macro frame when approaching like generative ai i think that's interesting is a lot of people attempt to try and make it look as real and as as as close to what real life footage would look like as possible and i almost feel like that's the incorrect approach a lot of the time and that's obviously what like arcades these ugc platforms are doing but I think the video that that you shared initially that kind of when you were first put the first concept you put together the space goods video which I actually shared around a lot of my group chats and like it was interesting just to see like normal people who aren't in the industry's reaction to the video they were like it just blew them away it's like jaw-dropping sort of like watershed moment from their point of view of like what these AI models are capable of from a video from a content generation perspective so i think it's just it's so valuable when you approach it with the right mindset and the right technical like skill set to get the right output from them from the models yeah a hundred percent i think that's definitely a massive learning from it like lean into the ridiculous um but then i think a few examples i'll go through is like there are cases where realism also does work and taking learnings from an account that will apply to a hook variation um you know the the visual maybe that has worked in the account previously how can we take that visual and 10x it in just craziness that's something we couldn't do before and the example i'm going to go through today is that insight that we spoke about in the last episode about the powder on the spoon i have now done a load of hook variations where it's like you know taking the learning of pov powder on a spoon applying that to someone on a roller coaster holding a spoon with the the creatine powder on it and you know all of these other crazy elements that that we are taking the learnings but applying it to a different setting it's like taking it to the extreme right exactly there is just just before i think it's key to note there are limitations we we were going to do a collagen ad where you have one person interviewing doing another and she comments on how great her skin is looking and you cut and she's literally a three-year-old child talking about her collagen routine but sadly google probably wisely don't actually allow you to generate babies talking about supplements they're not old enough to use it's a shame because that was one of your best ideas i think i think i think it's got legs eventually um but yeah let's dive in let's dive in yeah cool so i think let's just start like to get the creative juices flowing on some actual use cases some things stuff a mixture of stuff we've done internally as well as stuff i've seen externally um the first one is podcast ads so as we know this is a popular format on meta um works really well in in many ways and um you know we've started experimenting as well as a few others that i've seen taking the idea of a podcast ad but pushing it to the extreme so really good um segue there talking about the babies because the example i actually wanted to share was from motion um where they created an ad obviously with with babies on a podcast so let's give this a quick watch um ai queer strategies straight fire a hundred percent what's your favorite task to run in motion i like generate winning ugc scripts oh Oh, that slaps.

6:58That slaps. Baby, these AI agents, emotions marketplaces are so clutch. Yeah, yeah. Do you remember a few years ago when we didn't have all these AI career strategies? No, I don't. I wasn't even born yet. Neither were you. Oh, true. It's the mouths for me. I love that bit. But yeah, a great example of like, you know, taking something and pushing it to the extreme. but I think the key here is leaning into the ridiculous. And one thing that just is really important that I've noticed with VO3 is how do we, in terms of that creative philosophy, how do we lean into like satire and comedy?

7:39Loukas:And I think they've almost broken the fourth wall with that by then mentioning that they're babies and they're not actually old enough to have learned something. So it's a good example of that. Pointing out the obvious, right? I think again when you're looking at those examples if you're trying to make it like some people will be approaching it like let's try and make the podcast ad look like a podcast ad trust me it's usually easier just to film the podcast ad because the giveaway is still the mouths if you look at the mouths they're all over the shop when they're saying some stuff but when you're doing it in that context of like just daft concepts it kind of masks that you're not focusing on it you're kind of just focusing on the macro exactly video aren't you yeah for sure um i'm really interested just quickly by this example because i don't know if like google would have allowed this but the way just to follow on from your point earlier where if you're trying to get a baby to say something and it's not working the other alternative route is it will generate the visuals but it will have no audio so you take that and you then dub it with either an ai voice after or you record an ai you record a voice on top of it which i have a feeling is what they might have done in this scenario here it didn't look like it's synced perfectly yeah and also it's it can be difficult to get the voice to match up perfectly because you only have eight seconds essentially on on vo3 in terms of the output yeah um so yeah that's the first example the second one which is probably my my favorite of of all of them um is essentially like satirical skits and this falls into like tv c inspired ads that very lead very much lean into like satire sharp humor they're made to feel really big bold and you know they're very top of funnel now this is probably something we wouldn't be doing at scale within an ad account per se, but it's a good example of like how it is being used.

9:35And I don't know if you saw this, I think I sent it to you, but it's these guys called Too Short for Modelling. They're actually filmmakers, but they wanted to do a commercial for Liquid Death. And I think this is showing, you know, how the lines are becoming blurred with what, you know, the barrier to entry previously versus now.

9:54Loukas:They essentially were like, they love this brand and they've been trying to do an ad with them and pitch them an idea. but Liquid Death never got back to them. So they just went ahead and made a Liquid Death ad themselves. Have you seen this? I haven't, I don't think. It's actually incredible. I think it's one of the most funny ads I've ever seen. Okay, let's give it a watch. Good afternoon, son. Want to take a guess why I pulled you over? Oh, it's not what you think. It's Liquid Death. There's sparkling mountain water. Mmm, wow. You weren't kidding. That is refreshing. But it's not why I pulled you over.

10:29Oh boy. Is it the busted taillight? Nuh-uh. Is it because of the license plate? Nope. Shit, man. It's the dead guy, right? No, sir. Is it the human trafficking, perhaps? Nuh-uh. Man, is it the truck thing? I don't think so. Well, what is it then? The robbery? No. The bumper sticker? No. It's the roadkill, isn't it? I haven't heard of it. The chemical waste? Nuh-uh. Is it the rollercoaster seat? Is it the cult thing? Absolutely not. Hell no. Alright, I give up. What is it? It's because it's your birthday. Happy birthday, Kevin. Dad, you remembered. Happy birthday to you. You're the best dad ever.

11:07So stupid. So good. But yeah, just again, leaning into the ridiculous there. And I think if you're using it, you know, to do something like that, you really have to.

11:23Loukas:But a good example of you wouldn't do it to that level necessarily. but it's like how can you take elements of that and apply that to a paid creative yeah or take maybe you know do something like that within a hook kind of like motion did before we introduced the solution yeah love it just gives you just empowers creativity right like that was so such a stupid idea yeah yeah exactly um the next example i wanted to share was almost like that story driven montage um and i think the one that i wanted to share for this was the one that i did for space goods um and i think with this creative i just wanted to like really push it to the max so again it's probably not the best example of like we can do this at scale at least right now um it probably took a couple of days to to complete this maybe like eight nine hours in total which isn't sustainable like to be doing that at scale as we know but um it's a great example of like how you can again use lots of really attention grabbing obscure visuals while still aligning with the um you know the the rules of meta ads and the rules of like what a standard framework would look like through paid so we'll give this quick watch as well i think just brain is what a lot of people are battling, and that's what I was battling, and it's definitely held with that.

12:50Have you tried any functional mushroom coffees? Uh, like closed cup or chestnut kind of thing? Your mum must be so proud. I could attribute this trend to the perceived health benefits of functional mushrooms. Are you supplementing with functional mushrooms? Yeah, I drink rainbow dust by Space Goods. Combined with adaptogens and nootropics, it's designed to support your microbiome. Your energy and your overall mood. Your gut controls way more than just your poop, you know that? We're talking mood, immune system, Even your brain. Rainbow dust could help restore balance to your gut in just a few weeks.

13:20I've never had so much energy. I feel alive! Dude, coffee is so 2023. My go-to is double frappe, seven pumps vanilla. It's gotta be rainbow dust for me. Cleaner energy without the coffee crash. Bet you meditate too, don't you? Space Goods. Space Goods. Seven functional mushrooms, adaptogens, prebiotics, and no crash like coffee. Just pure, clean, cosmic clarity. so yeah just another example then i think it's a good way of showing how you can use a standard i guess you know unaware or problem aware story ad structure but still layer in the ridiculous of it you know so we're keeping the dynamism there because we're using different characters to tell the same story essentially but also the comedy is really important there because it almost works as the rehook so there's those moments that are a little bit ridiculous within that like you know a guy saying i bet you meditate too don't you or chestnuts um when talking about different types of mushrooms um so it's it shows like the importance of layering in those elements to vo3 creative yeah there's just so many cuts and scenes in there as well that would have when you look at it from a cost perspective how much that would have cost to shoot like with normal resource i also like the ability to align like the retro themes with the brand identity which is very difficult to do um so that congruency was really strong yeah it's a really good point that's

14:55Loukas:why i think space goods was a good one here because we could align their brand guidelines such as the pink neon vibe 70 80s vibe of the brand to the creative itself um which is something we'd really struggle to do otherwise um you know if we're filming that with art direction and everything else um and the final use case that i wanted to share um was doing hook variations so i think this is one thing that everyone watching this video can start doing tomorrow so essentially this is the idea of either taking a best performing creative that you have within one of your accounts right now or you know that you want to just continue to drive and um catch the wave and make sure that we're continuing to drive spend through the asset by doing further iterations.

15:43Loukas:Or if we've got a creative that is showing good promise, but maybe the hook isn't, the thumb stop ratio is sitting below 30%. So a real quick thing that we can do is test more hook variations as part of that. So this was an example that I wanted to share because it was aligned with what we spoke about in the last episode, where we talk about that creative that worked really well which was the pov first person shooter style where someone was holding a spoon and walking with the powder which was creatine on on top of it so i wanted to almost take the learnings from that and apply that to different types of creative and a mixture of taking unique learnings from the account and applying that to new creatives such as how do we put that spoon in different locations and wild locations but also just how can we create other maybe more realistic variations that could have quite easily just been filmed in our studio because i think you know it's worth it's important to test both you don't want to go too far ridiculous that you know you might be attracting attention but the wrong kind of attention and then you're getting high drop off um so i'll just show you a few examples here of some of the hook variations that I did for that concept that was already performing really well within the account.

17:02My mum said to me the other day that Create Team was her biggest hack during menopause. Did a bit of research and women... That's the first one. This is the second one. Again, this one was a learning from the account that we saw whiteboard concepts doing really well. So I essentially took that and thought, okay, how can we do some more of these through iterations? my mum said to me the other day that creatine was her biggest hack during menopause did a bit of research that white bod was vo3 generated yeah yeah and i'll talk through the prompt shortly to get these kind of outcomes this is the next one my mum said to me the other day that creatine was her biggest hack during menopause did a bit of reason this is another one so i tested this in both forms either as a split screen with the spoon that i mentioned to you guys uh with the original shot of the spoon below or i did this as a uh a version of just the spoon on the roller coaster my mom said to me the other day that creatine was her biggest hack during menopause so yeah a few examples there of like how in probably about 20 minutes i was able to generate

18:10Loukas:seven really strong hook variations of an already you know best performer in the account so powerful makes your shoots go so much further right like when you've got that body narrative or script or angle it just allows you to explore so many different ways to get it in front of new audiences yeah exactly so let's talk workflow like for those people listening how are we like thinking about operationalizing this or turning this into a workflow that produces like predictable high quality output yeah for sure so it depends if we're looking at this from the perspective of like a a net new idea or from hook variations i think just for the purpose of where we just were let's start with hook variations um so i'm gonna just share my screen now and we can talk through like that exact uh workflow so i've got essentially a prompt that is uh if we're if we're looking to to get hook ideas uh and brainstorm we want to be starting with chat gpt because chat gpt essentially is the brainstorming partner before we take anything into vo3 so the prompt is essentially you're a creative strategist leading ddc ad agency we'll pop it up on screen right now so i'm not going to go through it in depth um i need a list of vo3 hook ideas based on the following brand and concept information so that's where we input the brand name the product the angle the persona and the best thing to do to get really strong hook visuals the next thing we would do is uh input the actual concept description so you might already have that from an already best performer so you you can go back to to the original brief copy and paste the concept description and bring it into this same prompt which we're showing on screen right now the next thing it will say is generate three to five creative vo3 hook ideas designed for high performing short form etc etc the key here is then the the prompt will say keep the description fairly short and when we pick our favorite we will need it to be turned into the below json script format for vo3 but don't transfer it until we said so essentially then the prompt has the template of where what we want to put into vo3 so the key things you need to be thinking about that go in that json format which by the way um if we're just just so people are aware that's essentially javascripts as a text-based format that we use um which i found from split testing json works better than just like a standard prompt input but the prompt we'll be giving away down below remember to comment robot will have that exact structure that you need to be following but the key thing there is that what we need to be including as part of that is the camera so what camera are we using you know this is where it could say GoPro you know it could say like a really high-end camera it could even be the iPhone.

21:24ChatGVT will help brainstorm that with you as part of this process. Shot type. So is it an ultra wide shot? Is it a close up? Movement. So what camera movement are we seeing? Location. What location is it being filmed in and time of day? Because then it will really impact the lighting in the scene. Style. So again, this is where we discuss cinematic. Is it a doc style? was it a UGC? Scene and subject, so who is the character? We want to break that down in detail. The better you can describe the character, by the way, or the individual, the more likely if you're then generating a second prompt and you want that to follow up from the first.

22:05So let's say you've got a 20 second scene, you're going to have to break that into a few different outputs because obviously VO3 can only produce eight second videos. So what we need to make sure we're doing is tying the uh being very specific about what that character needs to look like just a little top tip on this to get that consistency with characters the key here is how can we anchor it or anchor it to something that we know the ai will understand what that looks like and what i mean from that is it's very easy to go on vo3 right now and type in do uh create a scene where of a guy uh holding a uh a corona bottle it knows corona if you think it's getting its data from

22:48Loukas:like youtube it's going to know very much what a corona bottle looks like and it can easily model that out similarly with if we want you know let's say i want him to look like kevin spacey uh but but change his eyes to this and this so he obviously doesn't look like kevin spacey it's going to be able to anchor it to something that looks very similar so that that's how you can get your consistency across the scenes if that makes sense that's a good hack um we then go environment then dialogue i think another tip for dialogue is if we want to really um exaggerate certain lines within the script so if i wanted to be shouting something suddenly but the tone before wasn't a shout do caps lock uh for the shout and because it helps the emphasis tone so what's the emotional tone is it excited uh deadpan etc um atmosphere again lighting um soft golden hour etc etc and then this is the most important one because one thing that you will know if you if you started testing v03 is it will use it sometimes just randomly has like gibberish that just appears on screen like just random text that just makes no sense almost like it's trying to create subtitles but it doesn't make any sense so as part of this prompt as with it within the constraints we need to say no subtitles no text no logos no music because otherwise it could add random text on screen it could add random music in the background so we need to make sure we cover ourselves in the constraints so essentially then we've got our prompt where we're going back and forth with the video ideas first so I've inputted my concept description already as part of this prompt that obviously we've shown on screen and then it's going to just give me a load of ideas so this one is like the energy switch no one talks about the three and then it will say what happens in in the hook idea it will give me a load of them and then if I say okay let's go with option you said number three or what usually would happen is you'd have to you know prompt engineering is all about re-prompting pushing for for more so i might say actually we're a bit too obscure there can we tone it down and do it a bit more like this um but eventually let's say i like one of the the ideas from chat gbt i'll go let's go with option three it will then straight away come back with me with the exact template for within json ready to copy and paste straight into vo3 so i'm going to then copy and paste that directly into vo3 click enter and it's going to then produce um the the output that that we need um so yeah that in terms of like brainstorming hooks that's that's the way that you can be utilizing chat gpt with it you mentioned a few limitations there are there any others to be wary of like are your best doing it in four second blocks are your best i feel like that's key for any ai right knowing so it's a support is as powerful knowing what it can't do as as knowing what it can because you can avoid a lot of like wasted time yeah it's a really good point um yeah so there's there's there's a few um i think one of the biggest ones is that if you're generating actual ideas you want to go from concept to script to shot list to prompt so when you as part of that prompt when you're asking it to then take a whole concept idea that might be 30 seconds it's really important as part of that if you're then working in chat gpt to develop it within your json format that you're very clear that they have to do each eight second block as a scene so then it breaks it up into eight second scenes because otherwise if you're not clear about that when you're going back and forth with chat gpt one thing that it's going to do is it you know it will then make the seat longer than eight seconds and it will just cut it off um within within vo3 it's kind of just that it's kind of thinking about how you would shoot it anyway right exactly you were trying to think about going through the same motions and processes if you were just executing on a shoot but you're just doing it on a laptop exactly exactly that um but just coming back to what we were saying about hooks the brainstorming like use as i say use chat gpt as a brainstorming partner and then use the exact template that we're providing you to then transfer that idea into a json script um as i say the key thing as part of that is like don't just take the json script at face value sometimes if i ask it to create a prompt it will just it will do that but like add random dialogue that i didn't want on chat gpt so just make sure it's as you want it however what i've actually just found best at this point in time because as we know chat gpt isn't the best for ideation is when we're generating hooks with vo3 it's like what are the learnings we can take from the account so for example the the what are some of the common visuals that we're seeing so let's say like we're seeing in one of the accounts i can think of one in particular i won't name them but you know we saw that external hooks were working really well like busy streets and a lot of you know motion and potential pattern interruption so taking that to the extreme i might just go to chat gpt and say create me a prop create me a prompt for vo3 using the below json um uh format that is a load of people crazily running in a circle on a street in london or like a tokyo walk crossway or middle of oxford circus at rush hour exactly so all you got to do is say that say but i don't want any hook dialogue and then it will take it and put it in the exact json format before you copy and paste that over to be o3 um so i've found that to work the best right now just in terms of taking those learnings or things that you know obscure visuals that we want to test and then it transferring that into a full script um a full prompt ready to then put into vo3 love it cool um one thing i did just want to also just touch on is just what i would recommend for how to use vo3 specifically so as you know we we need to use flow um which is as part of google Gemini um when we go to like create a new project we want to use text to video and then there are options for either VO3 fast or VO3 quality um now VO3 fast I would definitely recommend using over uh VO3 quality uh fast mode is probably about 10p per shot um and you get about 500 generations over the course of a month whereas quality I think you get about half the generations across a month and I have split test the two quality is slower and there isn't really much difference at all between the two of them so that's one thing just to be aware of I definitely recommend using VO3 fast over quality the next thing to just know is don't over complicate the prompt we've actually made clear in our prompt templates that we're sharing that you know that keep everything short and sweet and json structure helps do that because if you if you think that we're giving loads of huge amounts of detail is the best route like that actually confusing confuses things so keep your prompts you know no more than two paragraphs um you want to be very direct it's that idea of cut the fluff if it doesn't add to the story just get rid of it yeah i think you're asking the model to do a lot in that prompt right i think it's not like when you're trying to get it just to write some text or you're trying to output like some analysis it's like you're asking it to generate so many different parts that comes together to create one construct so it makes sense that it can kind of get lost if you go too far yeah um can you use okay could you use can you just put the prompt in and ask it to refine it for you i guess in vo3 well in like say you've got a prompt that you think is a bit too long could you literally input it and be like let's cut this down a bit more concise you wouldn't do it in vo3 more gpt yeah that's why like i really encourage this this workflow jumping between gpt and vo3 because what i i would get the prompt where it needs to be in gpt and then if i need it to if i need it to trim it down i would then re-prompt it in gpt yeah yeah exactly in there um i think so that's that's another thing that's really key um touched on the random subtitles thing so that's that's all good i think another thing that just for people in the uk that are listening to this i mentioned previously how you create that consistency between scenes um so there's a scene builder mode which allows you to add on from the previous prompt of where you just were it's not very good at remaining consistent so as i said the key things there if you want the character in the scene to be the same as previously be very specific about what you want them to look like anchor it to a potential celebrity or someone who we know is all over youtube because that's where it's getting a lot of its data from but the best way which sadly isn't available in the uk yet um because of all the legislation around ai and how much slower we are to get things which by the way is really annoying isn't it yeah it's crazy especially with this this has been the one case where it's been the most frustrating i think you can kind of deal with it but like just vo seeing all the americans jump on vo3 like way two three months ahead of time i know it's like the first thing where it's felt really frustrating yeah i actually listened to a podcast about this the other day and it's like um america the reason why ai will become so dangerous is because america will never put legislation on it because otherwise that will slow them down against in the race against china yeah yeah that's the see a lot of elon musk talking about like his like in ex descent existential dread around ai and like that acceleration and yeah it is it's like an arms race isn't it yeah exactly we're a bit europe's is similar like europe's also got very very strong legislation but it's like what's the point if you're like it's like what's the point if they're not going to exactly you literally just it needs to be behind it's like it's either yeah which is another completely another topic but um so the just quickly to finish that so that this hasn't come to the uk yet but it's worth noting especially if you're not inside the uk the best way to match a character across two different scenes or like two eight second videos is there's a feature called frame to video so let's say you generate a an output and it's one character a guy let's say an old man you want the next prompt to be the same man in the same house you can download that visual from vo3 you can record the screen um sorry you can screenshot that screen of him and then you can upload it to this thing called frames to video within vo3 and then you can continue to prompt say use the same character as this and then you go on with the rest of your prompt and that's how it like will link it to that same visual and that frame to video essentially you'd assume it's going to get to a place where it can just hold that context in a chat and it's just going to be able to just going to be able to flow through and just prompt a full feature-length film 100 % start to end yeah it definitely will like if you look at context windows of like what chat gpt used to be versus what it is now um it will definitely get to that point perfect so i think consistency is key for brands right i think is that especially if you look in we were on the performance lens where we're trying to be more abstract we're trying to be pushing the boundaries but from a brand team i think it's if you can get that consistency and congruency it can be super powerful there as well um which is probably even more important for sure i wanted to i think next on on the document that we're working down is like product integration and sizing like i've found this challenging before even on chat gpt image generation trying to get the products get it could get it in someone's hand without them having 19 fingers like all of that is like hands it still struggles with hands as well yeah hands and mouths is like where it just gets a bit lost but at times but yeah how are you approaching inserting the products into it's difficult we're not there yet um so there's a few ways that you can do it so one way is which we're not we can't get into too much depth today and i it is something i'm still really experimenting and playing with but um there's this thing called flux content context within black forest labs so if you look at something called black forest labs you can essentially upload a product render and it will then either put it in a scene for you that then you can take across into vo3 um but the way that you will be able to do it is from that frame to video so uploading a product render of that frame and then um building that into a scene but it's not yet at the point where it's strong enough in my eyes to be using it effectively is it good at pulling you mentioned corona could it pull a corona yeah i'm assuming that's just its context window like its context is youtube you assume and anything else that google can get its hands on yeah so you'd assume that brands that have that video video assets already out there if you're prompting a nike nike you're prompting corona definitely prompting ketchup like you get a big tier one brands like the the the household names will straight it knows yeah because it's there's just so much like you say the context window for those products is massive and there's nowhere currently to upload localized content context in a way that you can with chat gpt or to the same level you can with chat gpt so you couldn't take like a internal feature like loads of your content that you've made and dump it in and then allow you to prompt from it yet not that i'm aware of no interesting you expect that to come it will yeah because i think what we're all seeing right now even with image gen on uh chat gpt is like it's the text normally that it doesn't get right like you can get the product and the shape there but not usually the text um and that's why i think it's like what are the techniques that we can use that don't necessarily require having a product there or like what emotion did where you're using that to really set up and agitate the before state and then when we get into the solution we are you know we then bring in other other content and that's where like the ideas of fake podcasts obscure hook variations street interviews you all of these different things that we can really set things up before we actually then bring in shots of the b-roll of the product yeah or even like what you showed earlier with the with the supplement is just doing the powder just doing the capsule just doing the drink exactly it's like that's the thing it did a really good job on the space goods of like holding the brown powder on a spoon because you know if you can visualize that element without actually showing the branding then then that works well and the reason why i could get away with it in the space goods ad is because I actually just used image gen at times um so I was like snapping polaroid shots of you know people holding the product in the 80s um as a as a way around from like having been able to do that in the video makes sense interesting to see when they do bring that upload because I do think that will just make it so so powerful you imagine that agent by brand video that can just you can just dump all of your asset library in and then prompts away from there exactly crazy it's pretty mad um but yeah i think like in terms of the prompts that we're going to be sharing as part of this the the main workflows that like we're using this for right now is turning concepts or ideas into vo3 formatted prompts so that's the idea of using that exact prompt that i mentioned loading in a concept description and then helping it develop the whole scene from that direction so taking a brief to concept to a brief to concept to script etc brainstorming ad ideas as a whole um again you know you can do that in a way that um one thing that the workflow that i'm using quite a lot is like if i'm seeing a visual somewhere whether it's tiktok online maybe from a best performing hook if you want to like replicate that visual through vo3 the best thing you can do is upload that video into google ai studios and ask it to give you as detailed a visual description of what it sees on screen you can then get a really great representation of what that what the visual is from what google ai studio describes and apply that to your prompt for vo3 and i've actually tested it and it does it really well like it can describe the visual so well and then you can apply that to vo3 and it can recreate it in a very similar way mad i also i think obviously we've been talking about getting to a finished product but i also think it's power the power is also in like if you're struggling to translate an idea to someone else like the ability to get it like halfway there through vo3 and then providing that for them to take and maybe shoot or exactly translate into a final asset through that lens is really powerful yeah so i think this as i say the second one is brainstorming ad ideas yeah jump between gpt and vo3 lean into existing frameworks that you know are working for your brand say here's a framework that we're using um now i want you to help me come up with ad ideas for an obscure vo3 ad that uses street interviews as an example again we've got the prompt flows for you for that and the final one as i said is is the creating dynamic hooks and we've got the prompt flow that we'll be sharing for that as well um yeah was there anything else any other questions uh not for me i do think it's important to just state that there's some there's some there's some boundaries that people shouldn't cross in terms of copyright like yes i think um and it's quite interesting to see how this evolves because you saw what happened with the image gen and that um really famous anime yeah i just completely lost my mind like i don't want you mean i've seen a few really good actually on twitter um wes anderson films that are like brands have made wes anderson style content just using vo3 but the copyright is a big risk for brands or anything that's trademarked so because VO3 will be able to produce Mickey Mouse but there's no way you should definitely don't put that in your ad don't put that in your ad account don't put in Star Wars don't create Joe Rogan talking about your brand or high risk you could make someone like one removed from that but yeah just be careful with that i know some people will be pushing the boundaries um but yeah i guess my my only question to finish is like where do you think it'll end um now you've been deep in it like what do you think is because i think it's gonna i was reading the theory a while ago um i think it was it wasn't paul graham but it was someone like that that was talking about how it's going to get to the point where you can open netflix and it knows who you are yeah it knows your interests your behaviors your preferences what you've been watching for the last six months what you're really interested in now based on all of your personal data and it'll be able to generate a completely new feature length film in an instant series yeah directly related to your personal preference that counts for all of your interest likes and behaviors because of an algorithm because of an algorithm which is kind of scary man that's really scary but that's that is that's exactly it like that That is where it's going.

43:31And it is scary. And I think I went into VO3 a little bit lackluster because I went to film school and I'm a creative person. And I was like, oh, this is going to destroy the industry and destroy creativity. But by going through the process of creating that Space Goods ad, I genuinely was like, I've never felt more creative. like it almost allows you to really stretch your creative legs because the opportunities are like limitless yeah just empowers you to to takes off like the limitations or the shackles off of like what's possible and you can i think it yeah just what we mentioned about taking things to the extreme bringing in completely different like formats styles all of that just with a prompts is crazy but yeah interesting to see how it develops i'm just interested like with the legislation piece what happens there because do you have do you have to say you have to you don't have a watermark yet do you i know with those like those ugc characters i think you you do on you do on if it's something i actually don't know to be fair i'm not gonna i'm not gonna say something i'm not clear on i think i i know that when you're when it's got likeness to an individual i think that's where it's like a real area of like you you have to make it very clear so if it's a ugc character that's not real you have to make it very clear that it's ai generated i don't know what the legislation is for just a scene that has nobody in and is just generated rather than shot i'd argue that it's pretty hard to tell so like how are they going to stop that i guess that i think it's interesting that that opens the um going for a bit of a tangent but opens the door back to towards like um nfts and verification of content and um how that like whole thing that was very fad like and didn't really have that much utility becomes quite powerful when paired with with a technology like this really good point i think i think that is very dependent on the legislation piece because like it is obviously going to get to a point where

45:42Loukas:you wouldn't need to do a shoot in the studio filming customers if you can model that quite easily however if legislation means that you have to be very you have to make it very obvious that this was ai generated i think like covid and people working from home and craving being in the office i think it's going to create a craving for realism and things that are actually real um so i i can't think replacing that anytime soon yeah i think that this is gonna if you think about like facebook ads and d2c marketing like it's been the wild west since it started until probably the last two years in terms of what you can do until you've got these health category um event limitations like you can still pretty much say loose very very loose stuff on facebook and i I feel like this is, I do think the AI generation piece will drive like a real change in legislation.

46:42Yeah, it has to. Because otherwise it's like, there could be infinite amounts of just lies on the internet of like fake people selling products they've never consumed to an audience that doesn't know they're not real. And it could be a product that just doesn't do what it says on the tin, like we've discussed on the previous episode. So it's like, that's just a world that's the utility to customers and the utility society of that is just very negative, right? I guess the flip to that, though, being Devil's Avocado is like, what is the difference between an AI character talking about how good a product is versus paying someone and sending them a script to say the same thing?

47:24yeah that's what i mean like that should if you think of it from like a pure like outcome for the people like that shouldn't be acceptable yeah so i think that's what i mean i think it's not been legislated it's been good for business that that's not happened but there's going to become a point a watershed moment where that has to rebalance towards like authenticity and realism and like ugc creators have to be people who have used the products um they have to be like customers or all of that just like corrects itself a bit i feel like it's probably overdue that that's the case yeah so true um let's hope that day never comes yeah we make our job really hard um i'm excited to see what it does for like top of funnel marketing yeah like we're in a world where it's very it's going to help us a lot but like that the ad you showed earlier like i think it has so much power for like tv style exactly in a way that's even probably even more cost effective makes them so much more accessible as well in a way that just wasn't possible before i think yeah that's probably the one industry that in terms of production is

48:32Loukas:it's gonna especially like vfx i know i forgot what film it was but there's already a film that's utilize vo3 for like um graph um vfx that hadn't um that that if it would have cost them i think around 500 000 pound more if they if they hadn't done it through vo3 and then you know you look at that video that i showed like the liquid death one that would you wouldn't know would you really like that just looks like a great tv ad that is done with not even leaving the house yeah and i think we don't the fact that we don't really notice i could you can kind of tell when you're close to it but if you're putting that in front of the general public who some some of them still find it hard to transact on a website when you're taking it aggregate level like it's just not no one's going to notice everyone's going to think that's normal just yeah footage i think we can probably do a follow-up to this in the next like eight weeks because i just think it's it's just the law of accelerated returns like it's just going to keep developing at ridiculous rates 100 % I think this has been super valuable I've learned a lot I think people should be spending a lot of time on this and probably less time on arcads and these like influent UGC generation I think it's so hard to the amount of time you have to put in to get that out of that platform when you could just go pay a creator is just so much different to what you can get out of VO3 yeah and as I say I think as an action start with the hooks it's an easy way in there's lots you can do with it really quickly and you can just iterate on all of your best performers through some obscure hook variations um but as i say comment down below robot we will uh be sending this to everyone um that has has commented and subscribed make sure you are subscribed because we've got some banging episodes and guests coming up yeah i'm excited to keep these going nearly at 50 episodes big episode coming massive uh nearly a year of episodes yeah every week i think we missed one week because we're in south africa so we've been pretty consistently weekly now for a long time yeah yeah thank you everybody for watching cool thanks to the next episode

From the publisher

What happens when you give Loukas two weeks, a cleared calendar, and unrestricted access to Google’s new Veo 3?

You get a repeatable system for generating scroll-stopping ad hooks, without touching a camera.

In this episode of D2C Diaries, Loukas breaks down the exact prompt structure we now use within the agency group to turn ChatGPT JSON prompts into Veo 3-generated ads. Whether you’re scaling a best-performing hook or building a fresh creative system from scratch, this workflow will save your team hours and unlock new angles faster.

We cover:
 – How Loukas used JSON formatting to get better results from Veo 3
 – The key components of creating a high-performing AI hook
 – Real examples of hooks we’ve built and scaled across Meta
 – Why realism often kills performance (and absurdity wins).
 – How to use Veo 3 + ChatGPT to 10X creative volume in your ad account

We’re also giving away the exact JSON prompt template we now use in the agency.

👉 Drop robot in the YouTube comments and we’ll send it to you.

Watch the videos featured!
Motion's baby podcast: https://x.com/rezakhadjavi/status/1935690909984665644
Too short for Modeling Liquid Death Ad https://www.instagram.com/reel/DLkPcDmPchO/

This episode is for founders, CMOs, and creative strategists ready to build faster and scale smarter.

00:00 What is Veo 3 and why it matters for DTC creative
06:40 Examples of scroll-stopping AI ads
15:15 Examples of hook variations with VEO 3
18:30 How we use ChatGPT + Veo 3 together with JSON
25:30 Constraints to VEO 3
29:10 How to use Flow within VEO 3 
35:20 Product rendering + brand integration in AI videos
39:10 Use cases of VEO 3
41:30 Legal, IP and content authenticity concerns
49:40 Final thoughts + how to get the full prompt template

Follow Loukas on socials:
LinkedIn: https://shorturl.at/TRI8B
Twitter: https://x.com/LoukasHambi

Follow Olly on socials:
LinkedIn: https://shorturl.at/kPrqO
Twitter: https://x.com/oliverwhudson

Follow D2C Diaries on socials
TikTok: https://www.tiktok.com/@d2c.diaries
Instagram: https://www.instagram.com/d2c_diaries/

More from D2C Diaries

All 51 episodes
Scaling DTC Ad Creative with JSON Prompts and Veo 3D2C Diaries · 51 min
Listen in VO