--- title: "100% FREE AI Avatar With Gestures | Ultimate Open-Source Guide" source_url: https://youtu.be/3En1WtubzAY captured_at: 2025-09-24 author: "marat_ai" contributor: "ชิดชนก" platform: youtube --- # 100% FREE AI Avatar With Gestures | Ultimate Open-Source Guide youtube video by **marat_ai**, 7m28s, uploaded 2025-08-30. URL: https://youtu.be/3En1WtubzAY Tags: ai avatar, free ai avatar, ai video generator, livepotrait tutorial, cling ai, luma ai, eleven labs, open source ai, ai animation, diy ai avatar, ai face animation, ai body animation, ai image generation, ai tutorial, tech guide, digital avatar, ai tools, virtual avatar, ai video editing, digital identity, ai voice cloning, realistic ai avatar, ai character, ai for beginners, ai body movement, ai ## Shared in LINE Note+ - 2025-09-24 23:15 by ชิดชนก - 2025-09-30 22:11 by ชิดชนก — nearby chat: แยกตามประเภทครับ ## Description In this video, I'll show you how to make a high-quality AI avatar for free or at minimal cost. After extensive experimentation, I've developed a workflow that delivers impressive results without the hefty price tag typically associated with AI avatars. Throughout this tutorial, I'll guide you through each step of the process, from creating your initial image to voice changing. By the end, you'll have all the knowledge you need to create your own realistic AI avatar on a budget. Let's dive in and start bringing your AI avatar to life! How I made AI Avatar with Open-Source Tools | Ultimate Guide How I created AI Avatar with Open-Source Tools | Ultimate Guide Generate High-Quality AI Avatars for FREE on Your PC - Open Source Tutorial This tutorial shows how to create professional AI avatars using free open-source tools that deliver results comparable to expensive paid services. 🔝Key advantages: Completely free and open-source Works on modest hardware (8GB VRAM minimum) Quality comparable to paid services Beginner-friendly interface 🚀Performance benchmarks: 5-second video: approximately 10 minutes (RTX 3090 Ti) Faster generation possible with rented GPUs _________________________ ## Transcript (captions:auto:en-orig) This took a lot of effort to bring to you. Who would have thought you could generate an AI avatar of this quality for free right on your own PC. I've tested tons of different methods and finally found the one that gives the best results in both speed and quality. >> Recently, a significant number of different open-source models for generating AI avatars have been released. However, there was a major issue with them. These models required extremely powerful GPUs with a large amount of video memory. But now the situation has changed. We have an AI model tool that makes it possible to generate high quality AI avatars on relatively modest machines with comparatively short generation times. And I'm talking about the tool called multi-talker but precisely about app one to GP. And GP stands for GPU poor. It's definitely not in use for comfy UI users, but it's not about them because we here talking about instrument for simple people who don't like BDSM practices. All right, let's start with installation. To be honest, I've tried so many different method and majority of them is quite complicated and not so reliable. So the best option for you is to use specific app to install AI applications and this is Pinocchio. To download Pinocchio, go to pinocchio.co and just download the the application on your device depending on your operation system. You can be sure that it's very safe to use especially if you're using verified scripts. I strongly recommend it. So to download your application, press discover button in a pinocchio and in the search bar enter one and it will be sufficient to find our app. Press one 2.1 and here press install with binocular. Choose your name for your repository and press download. Well, it takes about uh an hour depending on your internet. And after that, the repository will be available for using. Well, after 5 minutes, I've got the message that installation completed. And I'm not sure that it's completed properly. We've got message here that it failed but okay let's try to to run it. We just started our application and here is one for GPU poor. That's very simple application. So don't worry about all that features here. The first thing you need to do just choose the model right here way multi talk fusion X and then you download the model. So let's test it out our application. For this um we need to choose text to video tab here. Then uh reference images. Choose inject only object or people. Add your AI avatar which you generated in any image generator you like. I'm using my old one this lady. And automatic removal of background if you need it. In our case we don't need removing of background. Choose the voice. One person speaking only or maybe two persons. It depends on your particular user case, but in our case for testing, one person will be definitely enough. So, let's choose one our voice. Let's choose our audio fragment. You can trim the audio right here. I need to trim it >> because this took a lot of effort to bring to you. >> All right. Now, we need to specify our prompt. In my case, it will be a speaking woman is looking into the camera and gesturing actively. Yes, you can even specify this or you can specify anything you like. So, let it be this one category there. There are several options with different resolutions, but I recommend they use uh the lowest one because I think you don't have too much powerful GPU for this. And next step, very important, number of frames. You need to specify this manually every time. In our case, the voice to follow is uh had duration of three minutes. So, it should be about 75 uh frames. So, let it be 77. Number of info steps, let it be just a 10, but you can once again experiment with that. Advanced mode if you need it also is here. I think you don't need that. The next step is configuration. You need to specify your performance parameters specifically for you GPU. In my case uh for example transformer data type should be BF16 because I have uh 390i which works better with BF-16. To me in majority of cases you can use default parameters for UG GPU at least in the beginning but then you can once again experiment with them. Uh let's start our video generation. Uh here is a parameters which I used. The attention mode is auto. Py touch compilation is on because I have uh VSSL. You need to VSSL for that. If you don't know what is VSSL, it doesn't matter. Data type FP16 for some reason. Oh yes, I forgot to apply the settings. Apply changes. Now let's just press generate button. And now we starting to making our video. And all process could be controlled right here in a terminal. We've got our lock here. And also important information when you just started your first generation. It takes a lot of time because you need to download all needed models and I think it takes about 1 hour for that. But I already have my models here. I just can wait to generate my video. And for me, if we're talking about 3 seconds, it takes about five up to 7 minutes to generate it on my GPU which is uh 390 Ti. Unfortunately, I don't have a lot of ROM. Might be a reason for that long time, but at least it worked. If you don't have powerful GPU, it's still going to work on your machine because it takes only 8 gigs of VROM. But also a viable option is to use rented GPU. You can use any service for that reason. There are a lot of them like Westi, Saloti, all of them work perfectly. if you need me to make video for that, how to use them specifically for generating AI avatars because to me, uh, renting a 1490i for generating video is a really good idea. It allows you to save a lot of time, a lot of money, and is better than using paid services. At the same time, it gives you much more freedom. All right, after 8 minutes, we've got our video, and let's have a look >> because this took a lot of effort to bring to you. Well, as you can see, that looks very, very impressive. And to be honest, it looks much better than majority of paid services. But of course, it took a lot of time to generate. In my case, it's 8 minutes for 3 seconds, but uh to my experience, the video with 5 seconds took me about 10 minutes. A longer video would be better. And the limit of the length of the video is just uh 30 seconds. It's very important. And these uh videos will be chopped on chunks and then united after the processing of video. So you need to generate several videos at the same time. But it's not a problem because here you can find a quee management. Very very convenient. So thanks for watching. Bye-bye.