#AnythingLLM #TextGEN #LMstudio #Ollama #Claude #ComfyUI #A1111 #Forge

hi welcome I'm impact frames in this video I'm going to show you how to use if proom maker inside confy UI which is a tool to use local llms and also apis to work with llms help you to make workflows and better the idea is to make videos talking avatars music chat all of that so it's going to be like work in progress working all the time on this and it's going to be a continuous effort to make tools that we can use to a Ai and make better images workflows have fun and all of that so you can see here that we are using lava which is a multimodal model to make a Hu about this image about this girls like fighting an octopus in the space and you can see in the dep of space three woman there in Brave in vibrant suits they explor and that's the high cool image about this thing but you can also go into entropic or open gp4 and with the vision model you can ask for example hio is also a vision model you can ask to make this prom and you can see here that Cosmic Adventure alien Warriors Clash vibrant wild dance so you can see like it change this prompt it change like the reply and you can ask anything about the image because this is a chat prom you can have like different General port purpose right so you have different profiles here I made one that is Cortana I have Yuka which is a girlfriend Dr Claudia which is a doctor dilia is a hacker miss Emily is like a teacher for like Basic School to explain like student topics like in a on very easy way and then you have the prop maker and the pro maker image that takes an image as a prompt or you can use non to use like the default setting of The Mod and that's how you can use it and then you have like a different settings here these ones are for the local mes for all L you can use temperature or seed and if you keep alive the M it will keep the the model on memory and you can use this when you have a lot of memory and you just want to be faster right otherwise it will flush the model out and then it will load it again it will take a bit more time and all of that then we have and Whisper speech and we can enable it by muting it with crl M so I press contrl m in there and you can use like a text in here you can have the text area and put something in the text or you can right click and convert the text input to an input right so you can plug that into beard and then you will take this H Cosmic Adventure alien Warriors or whatever you can plug it straight from the response to if you don't want to see the response but you can plug it like this and yeah so you you plug it like this and now this thing will come into here and you can use that to make an audio right so in the audio you have like a different settings this is for the chunks if you have a long audio it will chunk differently with the overlap a bit less overlap if you want but this is like the default setting that is good torch compile doesn't work on my computer but it might work on yours if you enable this it will probably make it faster and then you have the speakers right if you use none it will use like a default speaker and if you use one of these for example the teacher Tre is something that I grabb from the Internet it's a voice that I went ahead and grabb from the internet you can put it in the confi whisper speech notes so you grab that voice and he will train it on the fly so it's a zero short training that he will do here with this voice and yeah so let me see how is this voice maybe I should have just okay so that's the voice how it sounds and it trains that on the Fly and you can use like a fem OG file that is about 16k sample rate and it will train it on that so after that I can push that into the dream talk node and the dream talk node is is a note for making avatars and it will make a video Avatar toer there so I put an image this image has to be a closeup of the face and then it has to be like an Square image like 512 by 512 or 1024 by 1024 but it's important to be as best as you can and that is like face right so it can be like a 3/4 view or frontal view but it has to be a face and then you can select like different if it's a male you select the male or the woman you select the W and then you have a neutral or any like an angry expression you have different expressions in here and then you can go ahead and just run the workflow if you run this workflow it will give you something like if it's a 20 second voice it will take like 60 or 40 seconds to complete depending on your machine if you have an a slow machine it will take longer and if you have a fast processor or a fast GPU it will take a lot less so we can come in here and see like one of the results of this we can go into confy and then look in the outputs drain talk and we can see a previous result wisdom in their eyes gentle Giants roam at dawn memories REM me we can for now we can try to run one of this to show how how it works and see if it's we want to do it with anthropic [Music] [Music] you can see also like the reply here what is happening in the notes in the log giving you a new alien creators battling in Cosmic F Galactic wonder and now is uh passing the MS for The Whisper of speech [Music] and I will add a bar uh TTS also and some other notes I plan to add here for making audio for movies and stuff but it will take a [Music] while okay now the drain to is kicking in [Music] [Music] [Music] [Music] okay finished now so that that was live how fast it is and now we can go in here and check it up in the new folder that just opened there Hau inspired by the vibrant sci-fi image Kian creatures Back Lane in Cosmic hu Galactic Wonder so the the response was a bit longer it has it made include this thing so include instead of just the hu included this but that's because I I put this profile here right so the Cortana one but you can put like none and probably will be a better response from hiu on this one uh particularly right also these custom profiles you can access them um if you go in here again we go into confy and then we go into C custom nodes and you can also go into e AI tools and in here I have a lot of presets that you can use there are the embellishments negative prompts profiles so we go on the profile we open it with Visual Studio code or with I recommend Sublime and there is a Sublime code editor that you can use and then come in here and you can make your own you can even see my host some characters that I made actually if you see the diaa one that's a super character that I'm releasing here for free maybe even more valuable than the noes itself and this one it teach you how to make like for making code right yeah I going to try to include as much as possible as I'm making it and yeah that's it that's how you use drain talk now we have parler TTS parler TTS is similar to whisper as I said I'm going to make a bar TTS I'm in the process of doing it is a bit complex bar TTS is going to be a lot faster as you can see here the most that that took time was the whisper speech because I don't have the torch compile in my machine I cannot use this I don't know why at here we can see contrl M and basically we did the same thing we converted this into an input and now we can come and instead of training a voice with this the differ is that parer TTS is amazing because you can give it a prompt on how you want that voice to sound and if you give it like a voice like a sensual voice or a female speaker and stuff you will take that into consideration then you can go ahead and plug it back in here and you can make like chat Avatar with that voice that exactly the same that we did with whisper speech I don't going to show it right now because it's taking like a long time in the video you can see like this video took like one minute to make that voice and so the voice will depend on how long is the text and now we can see the other the other notes right let me just contrl m this okay so we have a prom to prom node which is for making just promts and if you see my previous video about this you can use it for making prompts oh we can come in to all Lama here and select yeah this thing we can select the model that I have if if AI is the prom maker and you can do like a light Warrior right so we can UE this [Music] prom and you have your Majestic Light Warrior standing to that D the M that I just released this as you can see it's called Lama 3 if AI is the pr maker this is a mold that I made yesterday so lama lama came out actually before yesterday like lama lama came out late Friday and I made this Saturday and this one is made with 50k synthetic data set that I made especially with a stable diffusion Proms and the idea is that it will help you to improve your promts like it because lama lama 3 has a lot better understanding of the context of the input that you put and it has better reasoning it's able to construct better Proms and on top of that I did like this 50k data set that is telling information about the lighting the quality of the prom so it will have like a better Aesthetics and it will be less wor salad and more useful words that you can use in the actual prompt and if you have I haven't tried sd3 but if you have sd3 it probably will give you even better result because sd3 is able to understand the words and actually use them in the pront right used them in the image I don't I don't remember the name of that how is the name but basically has like a template that I made especially for this and if you haven't used as I said you can just install it with one click if you have something like anything llm that I was just showing anything llm can run all Lama 3 and if you stall anything llm it will come with o Lama and then you can just download your model select them from the menu I think I already explained that on the last video Yeah so basically if you want to run this mode install it in your computer you just come in here in the link in the description I will put it in there and to install anything on all Lama you just run all Lama run and then the name of the model and if you do that in no Lama here let me see let me just put something that is better okay so we have it here and you can see that ama is a CLI tool you have like AMA hphone hphone help and it tells you like the different commands you have the wrong and the wrong model the name and the first time you do this this will install the thing in your machine but if not you will have like if you already have it like me you can just start making prompts for the advantage of this as I was saying for example B on crisis AMA before and many other mods didn't knew what bubblegun crisis is and bubblegun crisis wasn't even in the data set for Lama but now he knows what even the characters are he knows the the s name of the character he knows he in up the setting the apocalyptic Tokyo maybe it could Boomers and all of that okay so it has like an understanding of the thing for example when I come here and I ask pH princess he knows that am going for a fantasy scene so it tells a f fantasy scene a fairy tale and it takes glowing mushroom all twin clean fireflies so you can see like it's putting this visual representation into image and it will give you like a better image if you have a decent mold like protus that is a super nice mold and even another ones you will find like a super amazing uh image at the end so if you have something in here like antique showroom or something like that [Music] can see it working and it will come and it will give you a response like by quickly because it's now loading the model again because it passed more than five minutes and then it has to load it but you can see like the the response is instant yeah that's the amazing thing of working with this so now let's go ahead and show you where to get all these models all these things because you will have to probably some of them you can go into confy UI you can go into a manager inst start custom noes look for if Ai and you can see whisper speech and E AI tools right but the other ones are not present yet so you can come in here in in my tools in E you can see we have dren talk you can just click on that and please uh start the repo if you think this is worth it and it tells you what you need to do so basically for you're going to need FFM Peg and it tells you how to install it you're going to need to put the environment settings have the things and the description how to do it everything exactly how you have to do it and if you are in Linux this is how you do that in Linux how you add FFM peg into the path you first so app update so app install FFM Peg and then you put it into your pad you download the models put them into this place into your confi mods D all checkpoints and that's it so you have your models in there and then you will be able to use drol for the other nodes and again you just have to copy this into your thing so let's let's go for example parlor TTS so you can copy this right and then go into your uh let's go into to the windows side Windows machine we can go into confi and then we have custom nodes and in here you can just actually I have it here in this command you have to move you have to change H because mine is in ag but you just have to change that for C or D depending of the place where your machine is if you are installing this on a separate environment then you have to just activate the environment and do pep install G uh parer TTS so you have to install that in your machine and yeah so we are here in the paror TTS thing and we can do CMD as I was saying CMD here and now we are in there and now you can paste that and that will install with the portable environment for python the embedded environment it will install this because I already have installed probably will say it's already install and I don't need it and that's the way you install things with the portable version so you can if you have something that you like only one thing to install you can use it this way see what else we have if paror TTS whisper speech we can uh install it with the command in the inside the custom the manager yeah that's the way you do it the everything is like explain in here how to do all these things and that's it I think that's all for this video I think it's quite long video thank you for watching and I hope you enjoy this thank you bye-bye

Channel: ImpactFrames

Views: 1,792

Keywords: #LLAMA3 #Stablediffusion #AnythingLLM #TextGEN #LMstudio #Ollama #Claude #ComfyUI #A1111 #Forge

Id: 6ZkjkIJYR8s

Length: 19min 59sec (1199 seconds)

Published: Tue Apr 23 2024