Veo 3 MASTERCLASS - Character Consistency - Gem Ecosystem - "Hybrid Prompt Theory"

0:00 V3 is expensive. The daily limits come fast. Stop wasting credits by making bad prompts. Today's lesson follows directly from yesterday's lesson. I highly recommend watching that first if you have not seen it because it assumes that you've already created these two gems that we are going to be working with today. And I'm going to show you how we optimize these gems to make our prompts even better than they were before. Hold on to your butts, everybody. This is going to be the biggest episode I've made so far. And with this episode in combination with the last one, your prompts will be revolutionized. Let's get started. So, first just a recap. I asked the gem to give me an overview of who it is, what it does, and the best way to interact with it. Uh, this reminds me that here number five is the handoff and that this gem is designed to

0:47 coordinate with another gem. That is the scene weaver and the dream weaver are supposed to work together. So, I go ahead and ask the Scene Weaver who it is and what it does, and it gives me a much different kind of response, a lot more technical, a lot more broken down aspect of it because it is the specialized AI technical director. So, I expected it to give me a much more technical answer. These are things we set up in the last episode when we created these gems. So, that's why I strongly recommend if you if you don't know what I'm talking about, go watch that episode first, come back here and uh you'll be completely caught up. So today's topic was just simply going to be how to integrate both these gems and get them working on a project together coordinated very well. But unfortunately I was not able

1:33 to I was not able to show you that without uh letting you in on a few other secrets in today's episode. And uh three topics ended up getting covered today in once. So, we are going to be talking about the ecosystem between gems, but we also are going to be giving some tips for how to uh maintain character consistency between your prompts. Uh, and then lastly, I am going to finish off by showing you uh well, we'll get to there. It is it's a it's a good one. I'll just say that. But here it is. Uh, I'm asking for uh, you know, just the the specifics of its gems uh, capabilities. Uh, the hybrid prompt. See, I I knew I was going to get into it soon enough, but I'm not going to explain it just yet. But we

2:19 have developed myself and uh, this AI, we have developed a type of format for prompt giving into VO3 that is just gorgeous. But we will get to that in a little bit. You know, it's a it's a good part of it. this I'm telling you this episode is chalk full of good stuff. So lastly, uh it's answering the best way to interact with me. Whenever you're dealing with a gem, especially for the first time, this is a great question to ask is like what's the best way to interact with it? And it will give you it'll give you some uh things um to consider and you know to uh expand upon. And uh the the the last part, the most important part is that its final output will be a single complete readytouse video generation prompt formatted and

3:06 optimized for the AI here. I probably should have specified V3. Uh and if you're not using V3, I guess you can specify whichever program you are using, but nonetheless, let's go ahead and get started with Dreamweaver. This is our first uh conversation with the gem, and this is going to be where we work on our project. I'm hoping to have a channel trailer made for this using all AI clips demonstrating the system that I have developed. So that is going to be the specific goal for this conversation. As always, you can slow down the video or pause it at any moment if you want to fully read the conversation at hand. But we are going to kind of move through here into the important things. Uh I give it the general idea of what the channel is and then I tell it that uh

3:51 I'm also hoping to give paid consultations for fully complete and immediate training. hit me up. Emails in the description. Um, but I also would like to produce as many clips uh as possible keeping in mind VO3's daily limit. And this is, you know, hearkens back to the fact that we taught both Scenew Weaver and Dreamw Weaver to understand the limitations of the program. So, it knows that we're limited to 8second clips and so every clip that is going to give us is going to have to fall within those categories. So, here is Dreamweaver or Yeah, Dreamw Weaver. Got to not get mixed up with that. Dreamweaver, the one who weaves the dream the the visuals of the sequence, is now giving me the first step in the process and they're just delivering three ideas. I could talk to Dreamweaver

4:39 and I could be like, I have an idea for this. I want this character looking like this. I want this action. This is what said I can tell it everything I want and it can help me visualize it. Or I can do exactly like I said right here is I don't know what I want. you know, I I want a video, you know, I want a channel channel trailer using clips. I want it to explain what my channel is, but you know, I wasn't prepared for this. This isn't a project that like, oh, I conceived of and I have passion about. I want to this is just simply something I need to get done for work and and how can I get the AI, how can I get uh the prompts to work for me? So, uh I had it give me three ideas. I like uh I liked one of them. Uh, so we ended up going with what's called an uncanny picnic. I think it'd be an interesting idea to have a very stark contrast between a

5:25 very strikingly detailed visual of a person of a narrator backdropped on to kind of an uncanny valley of old AI technology. So that's the that's the the idea that we're going with. And so now it's um providing me with some uh with some more more options here. And I like that it gives the creative spark uh on each of them. That was an important addition to this gem's personality because it it's it's a little side extra thing, but it gives me the opportunity to see how it is thinking and gives me the opportunity to b to to expand on that thinking as well. But a question starts developing in my mind. So I ask it, is it possible for V3 to intentionally create poor quality imagery? and if so is it capable of

6:12 maintaining the realism of the character while also intentionally uh you know confusing the background image. This momentary quick aside that I thought ended up being a lot more important and defined the rest of the uh vision. I like to say vision instead of uh scene or video or things like that because it does kind of keep the AI in reminding that it's uh visual that you're trying to accomplish. But this ended up being a big thing because, you know, it it could probably render a very good human being narrator walking, talking, blinking, doing all that stuff. And it could probably do a pretty good background, but if I want it to be very specific with the background, if I want it to be the uncanny valley, do all that

6:58 stuff, um, that's a lot of processing power. That's a lot for Vio to take in. It's it's smart proc. It's a smart, you know, AI, but it's still limited. I'm not exactly sold on the idea. I feel like we still have a lot more that we can work with, but I kind of want to see how it's going to interact. What we've come up with so far, I want to see how it can deliver me uh a prompt that I can go and then bring over to the scene weaver to create the uh the realism I'm looking for. So, it expands on the first idea and then it gives me the handoff because I asked it uh for the command to hand it off. So now I have a little paragraph that I can copy and paste over into Dreamweaver and then or Scenew Weaver and then it can create the technical AI prompt necessary to create

7:45 this is the the conceptual vision. Uh Scenew Weaver is the technical. So a little sneak peek at the hybrid prompt theory is this sentence right here asking whether or not it should be developed uh whether the handoff should be in a codelock form or description form. Uh, and if you weren't aware, V3 handles codelock. I mean, Gemini handles codelock. It It's all It's all made to work together. So, codelock is an option. Long story short, though, the handoff doesn't need to be in Codelock that Dreamweaver and Scene Weaver understand each other well enough that they can just communicate in a single paragraph. At this point, I asked the question, do you think there should be any narration or is it too early in the video? You know, it's establishing shot, but what what are its takes? And the

8:32 thing about Dreamweaver is we included the fact that it knows about film theory. It knows about you know pacing and things like that uh viewer attention rate and uh so it knows about these things. So I wanted to ask its you know opinion on it. And so it gave me uh a couple different paths. The power of silence of not having any narration in the beginning or the immediate guide of seeing it uh or hearing the narration immediately. Um I kind of like a little bit of both. I like the idea of a little bit of silence in the beginning, her taking in the scene behind her and then speaking. And so, uh, we're going to kind of explore that idea here. As you see, the the AI is coming up with some stuff. Are you seeing what I'm seeing? This is what I have to deal with was the line that it came up with. Ah, it's not

9:19 great, but we don't go with it, so it's fine. Um, but, you know, that's, you know, it's all part of the create creative collaborative process. Um, so yeah. So it's it's fine for now. So I ask can it get ready for the handoff? This is again the command that tells it to finalize the package. So again I can go ahead and copy paste this over into Scenew Weaver and uh and then be go uh go on from there. I am going to show you all the Scene Weaver parts that we deal with this. But first I'm just going to go through all that we did with Dreamweaver. Show you all the conceptual stuff and then we'll get into the the details. I point out that I like the fluid motion contrast parts uh right here. That contrast is stilted family in

10:04 the background uh is a good distinction between the two images. But I do want to specify that the narrator is in uh high contra is is in high detail because that is probably one of the most important parts about this clip is the juxaposition between you know V3 and other AI software or the imitation of other AI software. So that's why this is a part that we're starting to peak into right now into the character consistency that I mentioned before. And this is this is where I started to develop uh some of the practices. So as we can see here, it changes the description of her from a two a confident female guide rendered in perfect lifelike clarity and

10:50 highfidelity detail. Whereas if we uh scroll up a little earlier, we see that the previous prompt merely just said a confident female guide. uh it did specify the fluid motion contrast, but it doesn't actually specify the fact that she is supposed to be highly detailed. So, I like having this extra uh moment where we took aside to specify that to include it into the prompt uh and into the description of the character. Uh and then this is exactly when I decide we're going to need a consistent character. We're going to need the same narrator throughout every single scene and we need to nail it down now before we start making clips and just start getting a whole bunch of different people trying to build the same thing. So now I'm working with Dreamweaver. We are going to come up

11:35 with a character. The thing is I didn't really have an idea of who this was going to be. You know, again, this wasn't my vision. And I I have a couple visions and projects that I do want to work on, but unfortunately I can't uh can't sacrifice any of the daily credits without uh using using them on things like this. And I would much rather have y'all learn some of these uh techniques first and so I can make my art later. Uh I'd like to just help everyone start making their art sooner. I asked the Dreamw Weaver who the ideal candidate would be um just given, you know, sociology and psychology of people and whatnot. And it came up with a woman in her early 40s. Uh sharp, intelligent, discerning eyes, dark hair, and a clean cut, professional bob with a noticeable, elegant streak, uh moves with poised,

12:21 deliberate economy of motion. Expression is calm. Very interesting character that that they had created. I like the fashion choices. It all sounds great. I'm just going to say okay. Uh obviously, if you have a particular character in motion or in mind, you can specify it here or you can change the details as you wish. And with the character profile fully designed, Scenew Weaver goes ahead and gives the handoff that we can now copy and paste this into uh Dreamweaver, excuse me, Dreamweaver gives the handoff for Scene Weaver. Maybe if you're going to make this, maybe come up with a little more distinct names and you won't have the problems that I'm having right now. But as we see here, we have two separate paragraphs. One for scene description and one for the character profile. This is much different than the version it gave us before and is uh a lot more

13:06 detailed. Again, the Dreamweaver is not meant to give technical prompts. This is not what you're going to want to put into VO3. I mean, you could and it'll it'll do it'll do all right, but we have a lot more technical control that we can give. So, we're going to skip past. We rendered the first clip. I will show that to you later when we're at that part. But all you need to know now is is it wasn't exactly great. I was having some problems with the detailing in it. And I had the idea of maybe having two separate clips. We either needed to simplify what was going on in the background because having her in high detail and then the background being, you know, uncanny, it was a lot, it was a little too much. It just didn't produce well enough. Uh so my thought was having the background be its own

13:53 scene then just we render the way that we're supposed to render it and then the architect uh is rendered on a green screen at which case she can be rendered in the the highfidelity detail uh you know she'll have the full path movement she'll have full audio control and all we have to do is then composite the two clips on top of each other. Um, it gives us the benefits of of both options, good and bad. And uh, honestly, both myself and the gem kind of realized that the composite clip was probably going to be best. Uh, the AI's focus is singular, and you can throw, I'm going to show you, you can throw a lot of detail at V3. You can throw a lot of directions at

14:39 V3, and it will produce high quality results. But the more you do, the more it has to think about. And the more it has to think about, the more likely you have errors. But if you can separate some of those tasks and you know, it's it's not great. You have to you have to produce two clips for this for this intent. But with this, I can guarantee the control and the quality of both clips. So, we are going to go with the composite. We're going to have the narrator on a green screen. Side note, did you know that you can prompt your uh V3 clips to produce a green screen? Um, yeah, if you need to, yeah, if you need to do any kind of motion effects like that, you can include green screen as part of the process. So, very very interesting. So, now it's asking us about the narration, what kind of uh

15:28 approach it wants to take and one of the options that it gives is the internal monologue. And I point out here, we want to avoid the internal monologue approach at all costs. V3 excels at lip sync. We want to show that. We don't want to hide it. Other AI systems can't do lip-s sync. So, they use other AI voiceovers and then they compile them and and that's how you do it. And, you know, that's cool and all, but if we're going to be making a clip in V3, we want to highlight the fact that it's this technology. So, that's why we're going to go with um one of the other options for the narrator that has a little bit more uh pause in her in her tone cuz that's just kind of the way that I want to go. And so now uh it refineses the narration style. And then I realized I

16:15 don't even know what she's going to say. We hadn't even gotten that far. Like I mentioned, this is a clip that was not very well um planned out in in in any sense. you know, I didn't sit down and be like, "Okay, this is the vision that I want to have." But I did want to have something that I can show you that you can go through this whole process of creation. Um, and by the end of it, not by the end of this video, because unfortunately there are limited number of credits, but by the end of this um this program, I'm going to have a fully fleshed out channel trailer ready to go using this uh this system. So, we realize that I'm not exactly sure what she's going to say until I realize what the background is going to be. We hadn't even paid attention to what it is that she's going to be reacting to. So, once

17:00 again, we go into brainstorming mode. It gives me three different ideas of uh backgrounds that we can go. I go with the first one because uh one of the other options was a little too complicated. Then I messed up here. I stopped the response and replied. So, that's just weird thing that happened there. So that's what that is. But we go with the impossible fountain which is rather than a picnic and a family and all that stuff, we just have um a welldetailed rendered fountain with just wrong physics, you know, just to show that it's beautiful but it's wrong. And so um now it's looking at what it would say, what the what the narrator would say when looking at a fountain that is beautiful but wrong. And I really like this one. This perfectly encapsulates

17:47 what I'm trying to go for with my thing is the fact that V3 is expensive. The daily lips come fast and stop wasting your credits. Perfectly encapsulates exactly what I need. I love the fact that I had this kind of creative back and forth and process. And the longer you work with a gem, the longer you work with a particular conversation, the more it understands your vision. And I've I've been working with this one long enough that it it knows the the type of uh mood that I'm going for with this. So now I say I love narration 3, but what what kind of accent would she have? If you notice here, I'm asking the question rather than saying I want her to have this accent or that. And that is because if you've noticed, sometimes characters will look one way but sound differently

18:34 on VO3. And that is because it's basically the action is determined more along the lines of how the character looks than, you know, anything else. And so I'd be better off asking the AI based on what this character looks like, what kind of accent she would have. Um, and it gave me three options. A transatlantic uh accent, a standard British, and a clear North American. We are going to go with option three, but honestly, you can go for whatever. However, if your accent falls outside of what the visual uh representation of the character is, you're going to have to do a little bit of uh more audio work with that. But we can get into that when I deal with audio work in a later episode. But now we have

19:20 the final detail of her accent down. We have how she looks, her actions. She's uh all this information has now been integrated into the profile. So, it is all part and parcel to keep the character as consistent across all the scenes as possible. As if you notice, we are uh almost 20 minutes into the video. We still have yet to put a prompt into V3. And that is because you'll notice that a lot of time the best way to deal in V3 is to spend the least amount of time in V3. You want to spend your time in Gemini. You want to spend your time making your prompts, making your descriptions as perfect as possible. You can run the the simulations over and over in here. You can ask it if we're,

20:07 you know, missing anything. Is there anything we're not thinking of? Explore different concepts. Explore the likelihood of whether or not it would work. Have you ever asked uh your AI what's the likelihood of this prompt being understood? You should try sometime. It's interesting. You can get some interesting results. But um so now uh it wants to do clip B, but I want to focus more on character consistency. I tell it I only have one credit left for the day, so I need to nail it. And that's a great thing to tell it sometimes because it lets you it lets the gem know that you're on a little bit of a crunch. You know, you only have one more chance. You want it to be good. Uh and so it says that the concept needs to be both breathtaking and achievable. And I love the fact that he used the word achievable. So, uh, it jumps ahead to a

20:53 later part in the, uh, clip, not even to anything I asked for, but it realized this would be a good clip to focus on because my intention was getting, uh, character, uh, cohesion, making sure that it was the same character across multiple clips, not using flow, using just native VO3. I like the line that they wrote for her, getting it right on the first try. That's the system that falls exactly in line with my channel and what I'm trying to do. The goal for this is that I have developed a system, not just tips and tricks, not best practices. There is a system to getting the appropriate prompts and that is uh that is what I'm going for. So they uh tell me that this is both a powerful conclusion and highly achievable in a oneshot generation. Um

21:39 so I feel pretty pretty happy with this. They mentioned that it'd be in a city and I first start asking and then I realized, hey, I did some uh clips on Athens, ancient Athens a couple days ago. What about modern-day Athens? So, they said, "Oh, this is a fascinating choice. It's a brilliant idea." And then it comes in here with a little bit of metaphor. It says, "Uh, she is the architect and Athens is the cradle of architecture and organized thought. To have her look over a perfectly rendered city that blends ancient mastery with modern life is a beautiful metaphor for what you teach. Applying a foundational system to a powerful new technology." So, I was just like, I did a thing about Athens earlier. I thought that would be cool and it came out here and pulled metaphor out of it. And what can be said for that? But so

22:24 now it has uh the dreamw weaver has now created the scene for the scene weaver and uh now we can uh copy and paste it. So now this comes for a little this comes a little bit later in the night. I came back to this conversation. I was a bit tired. I was about to go to bed. Um, I didn't really want to work, you know, I didn't want to start diving into it and breaking down stuff, but I wanted it to give me an idea of where we're at now, what we got figured out, what the the image is going to look like, what the video is going to come out to be. So, it gives me the potential blueprint for the full trailer, and it gives me, you know, the clips, just a just a brief summary. It's not I'm not going to take this and bring it over into Scene Weaver, but it's just for me to kind of get an idea of and uh you know, you can

23:12 pause or slow down. It's It's pretty good. I'm going to work on this. We're going to do a little bit of tweaking. In the last uh clip, I'll show you how we changed that as well. But for the most part, I kind of like it, you know. Um and this is this is nice. At any point in time, if you ever need to get a refresher on what you've guys decided on or or where you should go from there, just ask. It's a great way of confirming if you need to open up a new tab and then just like keep where you are in this conversation, just open up a quick little assign, you can do that as well. So, clip five I didn't like at all. The minimalist office, you know, the I'm here to teach you the blueprint. Subscribe and stop guessing. It's time to build. That's That sounds very AI right there. I did not like that line. I did not like the fact that we were just

23:58 out in a cityscape and now we're in an office. Don't like it. So I asked that uh if they could change the last line to let me teach you the system and make your visions come to life because that is something that I I feel very strongly about with my channel is is your visions coming to life. You know a lot of people want to make little AI swamp and that's fine. It's easy. You can do that. This channel is to dedicate for AI artists to perfect their craft and this is how you do it. And so it changes up the the clip. It changes up um the the shot. Oh, that was the other thing is I wanted to have instead of the office, I wanted to remain on the uh Athens terrace, but to have a drone shot that pulls up and away from her as she says the last line. And

24:43 so it goes ahead and puts that all into place. Um and then uh and then now we are going to work on it in Dreamweaver. Right now, this little conversation that you see here, I was going to keep this chat separate from the video that I was going to record today. So, what I did was I asked it, "Hey, could you summarize the um the project that we're working on, give me a little project blueprint that I can take and put into the exact same chat." Like, I'm just going to take this from Dreamweaver and put it into Dreamweaver, but into a new chat instance. Uh I didn't end up needing to do that, but that is a feature that you can do. You can basically summarize the work that it's working on. So, now we are in Dreamweaver uh or in

25:30 Scene Weaver. Maybe I do need to change those things. So, now we are in Scene Weaver. We need to produce a scene. I wanted to do it quickly um because this is just a test. I was running out of time yesterday when I was doing this. But one of the things that I say is I'm hoping that we can use the prompt given on the first try. So, I take the prompt that uh Dreamw Weaver gave for clip one, a Sun Dappled Park, woman walking through um you know, the uncanny family and all that. Uh this was the first one that we worked on. uh earlier and then I told you that we had problems with it. I'll show you how that looked here. So, it gives us the structure drafting and this is our format for a code block style prompt output. You can kind of slow down and pause uh or just read here

26:16 for a second, but this is the format that it always does. is it's this long uh singular uh line that you just drag over and you copy and paste all of that into the V3 prompt box and it will do good. That is the code block style that I mentioned earlier. Just copy and paste. Um, and the refinement is uh here phase three because as you see the sound design says details needed and that's because we didn't uh in the dreamw weave we didn't include any of the audio details for that and so it's stopping us and asking us hey what are the audio details so it made me think about it and I realized oh that is the point when I went back in Dreamweaver and I'm like oh I don't know what the audio is are we is she narrating is she not that's where we

27:03 started having all that conversation So, as you see here, uh, Scene Weaver is like, "Hey, what's the audio here?" So, I stopped. I went back to the other gem, coordinated it. We spent a lot of time working on that, and then that's when we came out here. And as you see here, this is where we came out with the character consistency guide. So, now the prompt that we give Scene Weaver now includes uh her character profile. You know, she is she's an architect, short bob hair, um, you know, uh, North American accent. It doesn't include her ethnicity here because this was before we had created that part. But now it is revised the um prompt to now include all those aspects. The family and picnic scene are hyperrealistic and flawed medium shot tracking the architect smoothly. All that sound design um all of that is

27:52 included now. And so I say that's great. Let's generate the final output. It meant to be a test clip but all the elements are there. So, I want it to be good. I want it to not think of it as a test clip, think of it as a real. And so, at this point, this is when my gem is designed to ask me, would you like it in the structured code block form or would you like it in a descriptive paragraph form or would you like both? So, typically I would just say both. And this is where you'll get to see those two in action. So, the first one we have is a structured code block. Um, and I'm going to produce the same scene twice. I'm going to show you the difference between a structured code block and the descriptive paragraph. But first, we're going to copy and paste all of this and

28:37 make it into a clip. Some of you may be a little bit ahead of the lesson right now and realize where this is going, having the structured code block here and the descriptive paragraph here and then me mentioning something about a hybrid prompt theory earlier on in the video. So, if you're keeping up, good job. But anyways, we have two descriptive we have two ways to describe the same prompt to V3 and as we know if you you can put in the same set of words into V3 and achieve different results. You can you know change it around you can do all that. So having two different formats will ultimately produce two different styles. So here is the testing. So the first one that we're going to do is the code block

29:22 right here. Um, these are all the things that were in that gray box. It was just, you know, copied and paste. Let's watch. So, uh, yeah, not perfect, not great. And this was before we decided any audio cues, any narration or that. Um, uh, yeah, that's the code block. So, now we go for the descriptive paragraph. And this is what most people are used to writing into V3. This is probably what it wants, right? So, now we see more detail on the family, but she is actually in kind of worse quality. I don't know, there's something about the lighting or she was a lot more visually uh detailed there than here. And as we see, it's actually kind of a different woman. The the the hairstyle is similar

30:10 and the the gray streak is in the right place, but her facial structure is definitely different. Um, and then yeah, this family was just a static PNG right there. Whereas over here, it was uh moving, but it also didn't really feel uncanny. It just felt out of place, you know, felt bad, but not in the way that we were trying to go for. So, I mentioned that it's a very interesting test. The code block had detail on the architect, but very odd and unrealistic movement. And the descriptive paragraph had very natural movements, which was nice. I like that. but it lacked all audio and she wasn't crisp as uh she wasn't as crisp as detail as in the code block. And I pointed out that in both the background was awful. So I was hoping we could learn from the difference between the prompts and achieve the best output from both. And

30:56 it is this point that the AI starts to analysis what went there. Why did one work better than the other at any given point? and it notices that the the structured code block gives superior detail on the architect because it acts as the distinct labeled checklist focuses intently on filling every specific instruction resulting in highfidelity detail. The motion felt disconnected from the subject is because it had a sum of its parts kind of thing. It focused on the checklist of what needed to be there and not how it felt. Whereas the descriptive paragraph uh felt very natural. It flowed very well, but the character wasn't there because that checklist wasn't there. So, uh, and then just take a quick look at the background, but that's when we decide to

31:41 make it into two different clips. But here, here is where we devise a hybrid prompt, and that is half of it is in code uh code block form, and then half of it is in uh descriptive paragraph form. Uh it's going to ask us a couple checklists uh for things to fix. But as you see here, we're going to look at the hybrid prompt here in a second. Um this is where I take a second to ask, is it even possible? Like can we have the hybrid of two styles prompts, a code block, uh and a descriptive paragraph? Will it even help? Like is this even possible? And I feel that double-checking your your questions with it is always a great idea. you know, even if you're you're pretty sure if the details like I figured this would

32:27 probably work, but I wanted its opinion on it. And so now it gives me a whole reasoning, a list of how best to use them, how to to put together the hybrid theory, excuse me, the hybrid prompt theory of a code block and descriptive paragraph. So now it gives us this. We've determined exactly what it needs. Everything is good, but it's still not quite right. So now I have a structured code block and b descriptive paragraph but it's not easy to enter that. So now I say how would be the best way to copy and paste that as a single block of text into the v3 prompt box. So it thinks about it. The goal is to provide a single cohesive instruction. Um and uh and it's it needs to be read by the AI because once again whatever you put into

33:13 VO3 into that text box it's going to take as an instruction. it is going to take it literally. It just want it just you put it in there and it does what it does. Gemini, this is where you can be flexible. This is where you can have human language um and make things. So, what we want to do is create our entire vision and goal. Let Gemini uh understand what it is and then feed us back the specific language that V3 would need to. And it's not perfect. I'm not going to say that with this system you're just going to nail every single clip out of the park every single time, but look at all the detail in this. You know, we we've got the the descriptive paragraph that that paints the picture. We've got every little aspect of the

33:59 shot put together and thought about. Spoiler alert, we actually don't. I come up with another little uh addition here. In a few in a few minutes, I'll show you what we added to that. But um oh evenly I almost stopped over the evenly lit chroma chroma key green screen. Oh this find I didn't even notice it until I was recording this episode. But uh you're going to see in this next clip one of the things that was wrong and it was that word there evenly lit. And if I had noticed that I would have had a much better uh result. So let's take a look at that clip. Right. So, this is our first uh Oh, I after I made that the the last clip, I started to write. Oh, you know, here's what I started to write to

34:45 Scene Weaver to get it to fix, but I forgot that I was in Gemini and it started to build the clip. It started to write everything that I put and make it a clip. So, I I cancelled the the generation. I stopped it. Unfortunately, I'm pretty sure that counted against my credits. So, I lost the lost the clip there. So, be careful. But here is our first attempt at the hybrid prompt. It has both the checklist and the narrative descriptions. And here is what it came up with. But yeah, as you see here, it's not a green screen like the way I wanted. She's actually literally walking on a green screen. And that was because beautiful. Now, imagine how many credits it wasted to get that so wrong. And that's because of the the lighting uh notation that it took. Now, imagine how many credits it wasted to get that so

35:31 wrong. And it's kind of funny because it is beautiful. It's a great looking clip, but the green screen aspect just wasn't correct. And I could have just taken a second, read through the prompt a little bit more, and done that. The second thing I noticed is this is now a different woman than the first uh two clips. Um, so I take a second, I ask, was there something in the prompt that specified that? And if not, is there a way to specify it going forward? Um I the the whole goal of this uh thing is I want character consistency to be designed. Um I noticed the flaw that I didn't have her consist I didn't determine her ethnicity in her character profile and uh if that is something that we need to specify can we nail that down

36:17 now to make sure that the next clip that we make of her is indeed consistent. Um and so I pointed out three times in my my question about consistency. So they see that is absolutely paramount and they explain that what I experienced is a perfect example of how generative AI work. Uh the model was given a set of descriptions like sharp intelligent eyes, dark hair and a clean bob, minimalistic architectural style and it just uh kind of assumes it just gives nudges. It makes uh its own decisions on that. It's not an error but a demonstration of why it is important to be so uh this is explicit but I you know specific in your character consistency details. So now ethnicity has been added to the character profile which uh was an oversight on my part. Uh that could have

37:04 saved me some time if I had just uh included that in the original profile. But it just goes to show that you actually don't even need VO3 to uh start working on the ideas that you have in mind because as you see here, the more that you can think about before you even enter anything into that prompt box. The more that you can uh figure out, nail down, the better. So now it's it's ready to produce clip B. But I mentioned uh I only have the one credit left for the day. like I said into the dreamw weaver and I'd like to see the character consistency in action. I give it a little bit of positive reinforcement saying I know you can do it and I let it know that it is hard work now but it will make things easier in the future. Um oftent times uh this particular sentence has worked well with me. Sometimes I've gotten some good results

37:50 but that's a whole other subject for another time I'm going to talk about. But so it understands that the last credit is uh very important and we need to lock in the process to make sure that the next prompt we put in uh is consistent with the last one that we did. Um now that the character profile has been uh changed based on that last clip of her walking through the green screen. That should be the the architect that we see in motion. So right now it is just generally doing its thing and asking. We don't actually need to have Dreamweaver involved in the process at all. We can work in Scenew Weaver and and just describe what we want to, but uh you know by having them work together, it just makes the whole thing flow a lot more easily. So now we have

38:37 um our prompt for the final scene that was the one on the Athens terrace. And I like that it specifies that the absolute top priority here will be ensuring the architect's established profile is explicitly and meticulously defined to maintain perfect consistency within the character from clip A into clip C. So now it's going to draft the hybrid prompt um all in one box that I can just copy and paste uh and uh bring it through. Does say please review and I strongly do encourage to review. Uh, you know, I did mostly review these things, but again, like I mentioned, these are mainly just for proof of concept and I was trying to uh get it out a little quicker than uh I well apparently needed

39:22 because I wasn't able to finish the whole commercial yesterday. So, I'm going to be working on that today. Hey, my credits actually just renewed about 10 minutes ago. I'm going to get working on that pretty soon. But anyways, I do want to get the rest of this video out to you. We only have a little bit left. Uh, I noticed uh towards the very end that they did in uh include ambient city sounds. And the word ambient city sound or the phrase ambient city sound can be a devastating one. You do not want that in your prompts ever if you can. But uh uh I'll I'll talk I'll talk a whole lot more about that when I get into audio design in a future episode coming up soon. But at this point, I take a look through the prompt and I see that the

40:08 phrase consistent character isn't there. And I've seen a lot of people mention the tip that oh just add parenthesis consistent character Patrice and that will uh keep your characters consistent throughout the whole thing. So I ask I'm like hey is that is that a thing? You didn't include it. Did you not need to? Is it not useful? And so it says is it necessary? No. The detailed multi-point description we've created, blah blah blah, is the most powerful and effective way to ensure consistency. Is it helpful? Yeah, sure. Think of the phrase as a helpful reinforcement. Signals your primary intention. Um, so it's going to add a a phrase. It's not going to say consistent character, but it does add a phrase into it that um I don't know. I can't exactly say it did or did not help, but the fact that we uh that we

40:56 added the fact that it is she is indeed an established character. They didn't say consistent, they said established in the prompt. So, that's what I'm going to go with. Um, and then lastly, that question that I have right under there is uh should I include the title hybrid prompt for clip C character consistency test into the VO3 prompt or is that for my reference mostly? What they said is that uh uh it doesn't actually matter. Um the any line in the prompt that begins with the two backslashes is a comment made by the AI. So the VO3 actually won't read that and and it will just take it as uh as a title. There's no harm. We don't need it. We can avoid it, but it does keep everything organized for our benefit. And sometimes it's more for the human's benefit than for the AI. Um, and then so and then I

41:44 have uh a request to add into its actual coding uh to change around how it handles subtitles. If you are fast enough, you can read what that code was. But uh I'll show you much more and I'll give you that exact line of code that you need to add into all of your all of your gems that will keep you on point. But for now, let's go ahead and look at what one single try at the the prompt created for our con character consistency test that uh our character with a specific established appearance. See, that's the phrase that it chose. Um it has, you know, the camera style, the art, visual style, environment, sound design, everything. It's got the narrative description, uh and uh narration. So, here we go.

42:30 [Music] getting it right on the first try. That's the system. Ah, wonderful. So, now that little mouth open at the end, we can edit that out. You know, there's no point in changing around anything, but let let's take a quick look. All right. Character face, right? Wonderful. So, not exactly not exactly the same woman, but you know what? we could take a little bit more time and spend some time describing her face a little more if if our character profile had that kind of information. It could uh it could keep that information along multiple clips. But as you see here, I reach my video generation limit and uh it will have to wait until the following day. And that is why I feel that this

43:17 course is so important for you because every minute that y'all are making prompts that are just a paragraph long, that are just the first sentence off the top of your head and not using this program, that's all those credits that you're wasting. And I don't want you to do that. I want you to make good stuff. So, we are going to finish off with my scene weaver actually explaining the hybrid prompt theory uh to us. So you can either once again slow down the video or pause it and read as it goes. This is very interesting stuff. I will probably make this its own posts detailing this method because honestly I feel like this is a revolutionary style of prompting um and it should be adopted by more people and I feel like people

44:03 will get a lot better results if they could understand this. And so please thank you so much for watching this video this far. If you know AI artists, share it with them so they can start making better use out of their daily limits as well. If you want uh one-on-one training, if you want to learn the whole system all at once, there's actually a whole lot more that I have developed that I will have time to to get into on this channel. But uh hit me up. My email is in the description and I can help you out in your channel specifically. But look at this. This is actually this example. I asked it to just provide an example of the code block that it gives. And this is actually even more depth than what I had uh shown you earlier. This is a lot better. And so it it surprised me that it even gave that. And then at that

44:49 point I realized, hey, we could probably add a section just dealing with the lighting, you know, cuz lighting is important. Lighting is a very aspect and V3 has good lighting. I'm like, oh, but we'll deal with that later. But nope, it just was like, oh, you want a lighting section? I just updated into the core structure. There you go. lighting lighting from now on. As you see here, a whole section just all about how the lights move. So, this is uh I'm going to be perfecting this uh hybrid prompt. Um and uh eventually I will be able to provide you with a copy of it of the perfected version of it. But for now, I am going to leave us off on a question that I've asked it a couple times now that I want to reiterate to you. and that is how many character limits are in V3. A lot of people will tell you don't

45:36 put too much information into the prompt because it will overload the model. It won't understand. It'll try to do everything and then it will just end up doing nothing or do crap and that's no good. But I really wanted to know how much information could we put in there. And it says, well, there's no known fixed limit to report. The most effective is use a detailed well structured prompt that is as long as it needs to be to clarify your scene. The system is built to handle highly descriptive prompts, but this is the important part. You should feel confident in providing the level of detail we have been practicing. So, the AI told me and assured me that this system will work. Like I mentioned before, I'm working hard on perfecting this system and I have a lot of documentation that I am willing to share with you. If you'd like to learn the entire system in about an hour, I am

46:23 available for consultation. Hit me up. My email is listed in the description below. The system is complex, but it's not complicated. And anytime I see someone say that VO3 is so limited and hit and miss, I just want to shake them and say, just learn the system. And as you see here today, there is indeed better ways of prompting. There's better formats to prompt. And there are ways that you can utilize the tools around you to make your vision come to life. I've been the VOGemini Tutor. Go practice what you learned today, and I will see youall on the next one. Beautiful.