I Built My Own AI Video Avatar (No Avatar Subscription)

Hello everyone. Today I've been working on creating this kind of UGC avatars for your video. I first generated this video. I also used AI to record it to to scroll the web page to and to describe the the content that the video is currently showing. So this I I used Codex to generate it and it actually created quite a good video and and step-by-step view of then uh of this this idea in Trend Seeker and how does it work. Okay, let's play it then and and then I can talk about the avatar a bit more later. >> This idea was added on September 18th. The opportunity is specific. Help two people finish a conversation when their internet connection is unreliable. We will check the evidence before deciding what to build. >> So it talks talks about what this trend is about, what this business idea is about. And and I I didn't even have to create this transcript or anything. The AI just generated it for me. But in the beginning I created it without this avatar and and then it seems kind of boring AI generated. And and I hope when I add this avatar that narrates this topic for the listener, then it will be a little bit interesting or more interesting. Yeah, so I want to I just generated this but I I just want to give you an overview of how I did it because I think it looks pretty good. And and then you can copy my workflow and and create something similar. So we can first look at the >> Help two people finish a conversation. >> my my latest status with the avatar. It uses Codex to generate the 2D cartoons and then Blender to merge the the mouth movement, the head movement and and the any movements together into a single video and it takes an input of the transcript and then later outputs this video that you can then put in the original video as well. >> Let's investigate an offline conversation translator for travelers. This idea was added on September 18th. The opportunity is specific. Help two people finish a conversation when that >> is that we we actually generate uh like three three images for a single hand movement. And this this also is everything happens automatically. Codex generates this on the on the fly or like not on the fly in the in the background for me. And then later with a script it merges it together to this final merged uh avatar here. Yeah, but let's see then what are the stats here. We we currently generated six gestures as you can see on the bottom. I looked at this uh YouTube video from >> 13 years ago. Guesty transforms short-term rental management. We relentlessly push the bound >> BUT FREEZE IT, HOLLAND. STOP, HOLLAND. KEEP YOUR HANDS ON IT. >> OF THE BODY IS important >> The person whose hand is most on top is saying I'll be the boss. >> narrator uh what are the gestures and the body language that you can use to get attention from your listener. So, so, so, yeah. If you heard the framing, then you need to put hands in the middle. If you're offering something, then you need to put the palms palms up front and and this this gestures can give a a lot of I don't know, not nuance, but liveliness to your video that you otherwise don't get with with the with the normal screencast. Yeah, and and we we have six of these. And it actually it was quite funny that the AI actually generated added this count two gesture at the point when there was the offline two-way conversation because I I would normally guess that uh this is kind of like a peace sign, but here the AI put it in as counter for two, so that was interesting. Let's see. How how does it do it? Um it makes the 1080p footage and creates a reusable character. That's the main thing because we we use codecs to create these frames, and so it needs to have this very specific input to always generate the the the same face for our avatar and the same clothes. You can see that it has a little bit of kind of not correct things here on the right side, but that's uh that's that's fine. Here you can see as well that kind of has I don't know if it's on Blender or something else. So, it's not perfect, but if you look at the video, then it's uh in the bottom corner, so you almost don't even notice it. And we used then the Google chip voice that is perfectly fine for me. It It sounds great. And gestures are matched to the original audio. So, as I said before, when there's two in the conversation, then the avatar will show two. And then Python scripts will build animation. And and Blender use Blender to render the frames. FFmpeg to combine the frames, speeds, and website footage. Yeah, so that We used the FFmpeg CLI tool to merge to the the final product together. This all makes sense to me. And then later, yeah, we we compare all of these gestures and things together here in our website. In the Blender, we also added the like the hair moving animation to to make the character seem even more lively. Uh the lip syncing I added as well. I I'm not completely sure how how this works, but yeah, the mouth opens. Uh so, it's not complete lip syncing, but it generally opens and then moves the mouth when the character is speaking in the video. Let's see the cost as well. Blender, FM FFmpeg are free software. So, this is just when your local computer is is working. And the Google text generation is is actually almost free if you're if you're not completely uh like I don't know, building a lot of videos daily because 1 million free characters uh you get per month for free, yeah. And the characters, yeah, we we used our codec subscription. So, it's essentially included in our subscription. I used a $100 plan. No, no, I know I updated to $200 plan. And I've been using it for 2 days and I'm completely at the at the normal weekly limit, so I'm not over at all. And and that's about it. There is nothing more to it. I will I will play around with this a bit more. I I also implemented a script, so or like a workflow to generate new characters to my other products later. But this one is for the 10 seeker currently. It matches the kind of the the color scheme of the website. Kind of darker hair, blue is navy suit. But for my other products, for instance, beemax.cc, I might use another character. May maybe hair, some kind of pinkish character or or something else. And but with the script, then we we have our pose library existing and we can reuse this workflow to generate new characters for my other projects. And then I can spin up my UGC and marketing workflows and start generating and publishing these videos to both YouTube and also the portrait videos, I can use the FFmpeg again to generate different variants of this video. I hope this will be a a good pipeline for me and and I can test out the new marketing features. And yeah, that's it for this video. If you have any questions on how I generate that if if you would like to see the specific workflow or scripts, just let me know in the comments and I will I will publish it for you. If you have any questions or comments, just let them know as well. And thanks you thank you for watching this video and have a good day.