when people watch my documentary they all want to know "How did you make him so consistent?" I'm here with the star of my V3 documentary to show you how I generated consistent characters in V3 they're going to want to know how you did this part too yeah you're right all right this intro I did in Runway v3 currently doesn't support character references which is the easiest way to create consistent characters so I'm going to show you what I did and the foundation came from my previous video called Wonderful Dream which I created in Whisk and the character from that makes a cameo in this movie all right so this is Whisk and it's a tool from Google that lets you generate images and then you can generate videos from those images um this is what I made Wonderful Dream in so this was the final prompt that I used um to create one image of the character so I didn't use it as a reference image but I had to create an image of that character this is what I used so if we were using character reference images we would expect this image to be what the character in the video looks like but all we're doing really is taking these characters and then we're going to go over here to the subject and drag it over here and why are we doing that so it's going to analyze this image and it's going to give us a much more robust prompt so this is what it sees so this is kind of how the Google AI universe sees this image so this became the basis for my prompt i took all of it over to Gemini and I attached this photo and I said "Here's a prompt I wrote in Whisk." and I gave it my prompt and then it said "Here is the image description I received for this image." And I put that image description in there and then I said "I would like a detailed VO3 description of just the man that I can use in a template for building prompts where I try to place him in a consistent looking way." So don't worry about the wardrobe just focus on his face just to recap here's the image here's the prompt I gave here's the image description from Whisk and give me something to use consistently so I just asked "What do you think his name would be and how would you write a voice prompt for him in VO3?" Cuz I want to have a consistent voice all the time it thinks about it it says Aram and I was like "Okay cool." Like I'm not going to I'm not fuss with you let's call him Aram and then it gave me some options for his voice here's an example of the full prompt so it gives you the physical description of him and then when he speaks this is um how he talks and I wanted a certain style so I said "Please provide a core prompt for Aram core prompt for Aram's voice and a core prompt for a cinematic style shot in 35mm." This is for the visual or physical description of him this is what his voice would sound like slightly raspy and grally tinged with thoughtful curiosity all right now do a prompt in a high-end documentary style where he says "I used to wake up and worry about all of the things I needed to do that day." And it gave me this i used to wake up and worry about all of the things I needed to do that day so that's kind of it for consistent characters build a very detailed description of the character and use that in all of the prompts okay so the trick is using the same detailed character description in the prompt but how do you get rid of these subtitles the other big thing that people have a problem with are the captions so I heard on Discord that if you use Gemini to generate the videos you don't see the captions so I tried that in the first one there were no captions and I was excited and then I did it again and I got captions so that didn't work so I wanted to try an AI way to get rid of the captions so even if VO3 generated it with captions I can remove them with AI so the first one I tried was Runway's inpainting tool you've got a brush over here just go to the captions and just kind of wipe the brush over the captions and then they disappear so you can kind of in my opinion you can kind of see a little bit when it adjusts there but it's better than nothing like if you didn't know it was there you might not notice like if you didn't know to look at that spot you might not see it because you're watching him talk so that was okay but then I remembered that Cap Cut also has something similar um I use Cap Cut i think it's a great AI editor so in Cap Cut I just put the video on my editing timeline as I'm editing the video i selected it i went to video and I scrolled down here to where it says AI remove i checked that it takes a second to recognize the video i selected one of the brushes so I did the quick brush just kind of like moved it over and boom it did that so I put a little like box around it even then I said remove him and look at that like even in the collar it didn't like there's all of these lines in the collar and it it's pretty it's pretty smooth so I think that the cap cut one worked terrific okay and then lastly one of the problems I had with his voice was I think because he's described as an Armenian farmer it would give him different accents i had a lot of jobs my favorite ones were being a butcher and being a barber and I really liked the voices that I created in the documentary interview styles so I combined a few clips of him in order to get 10 seconds and exported just the audio then I brought that into 11 Labs and 11 Labs will let you clone a voice with only 10 seconds of audio i added the clip of him talking and generated a voice is called the Aram so I exported all of the audio where he's talking and brought that into 11 Labs and used their voice changer and what I was hoping would happen is that he would lose his accent and it would sound like the same character so that didn't really happen some of the videos had an accent to them and so I just typed in and did like text to speech for those until I found ones that were close enough to the same timing where I could just drop those in so it's kind of cheating I know in terms of like it's not the same generated audio but um it is generated audio and I think that having a consistent voice as well as a consistent face really helps sell it it's not that I dislike reality it's just there's more than one reality it used to be 100 credits for every video in V3 and now you can actually do the fast ones and some of the fast ones turned out just as good as the pro ones in my opinion sometimes too the fast ones gave me like really weird surreal things like this one was not what I wanted to do but it was kind of uh very bizarre feel like this one was done in the fast one this is one of my favorite shots in the whole thing and this was done on the fast mode if you're wondering where to change that you go here to the settings button and you can choose between three V3 fast which shows you it's 20 credits or V3 quality which is 100 so I hope this helps you create consistent characters in V3 and if you enjoy this kind of thing please subscribe because I like to make movies and I like to do tutorials about how I made those movies all right before I show your blooper reel do you have any final wisdom is this where I'm supposed to say smash that subscribe button i don't think people say that anymore all right let's watch some Bad Generations all right thank you for watching um if you're interested in any of the tools that I use today please check the description and I put links to different things in there all right go make some consistent characters in VO3 [Music] [Music]