Full transcript
0:00Hello everybody. In this video, I'm
0:01going to show you how you can create a
0:04local AI chart on your computer using
0:08just two pieces of software and how you
0:11can customize it very easily like you
0:14would do with any other charts that you
0:16tried online such as text generation web
0:19UI, silly tavern and similar. Now, I
0:23created this chat mainly for me and
0:26while I was creating it and developing
0:28it, I thought to record a
0:31behind-the-scene episodes, actually more
0:33than one, to share how I go through the
0:37process of creating something from
0:39scratch using LLMs. In this case, I use
0:41Quen Studio, but I also use Chart GPT as
0:44well. In this video, I'm going to show
0:45you the final work about the interface,
0:48how it works, and then I'm going to
0:50share with you the interface itself. You
0:53can just uh click on the Patreon link at
0:57the bottom or whatever links is
0:59available at the bottom of the
1:01description and you can have it as well.
1:05Okay. So, let's share the screen and
1:08let's have a look at this app. The first
1:11things you see when you open it is this
1:13welcome page. And you see this is called
1:16Aura AI and the version is 1.26 because
1:19I did roughly 26 or 27 iterations of
1:24this uh interface. This is an HTML file
1:28essentially and that's all it is. And
1:31what it does it allows you to chat
1:34locally with your AI. So you could do
1:36that using LM Studio. You can use that
1:39uh using any other local LLM, but this
1:43one also gives you the option to
1:45customize the voice. So you have a
1:48speechtoext feature and you also have a
1:51texttospech feature as well. So you can
1:53talk, you can converse with the AI and
1:56it will reply you back with a voice
1:58almost like having your own private
2:00Jarvis. In terms of responsiveness,
2:04it's going to be fast. if you load a
2:07small model or at least a model that
2:10fits your graphic card. So here you see
2:13all the things that you can do. So you
2:16can chat you can create charts and
2:18folders very similar to projects in chat
2:20GPT and in uh Quen Studio. And if you
2:23want to see it, if I go to my chat GPT,
2:26you'll see I've got all my list of chats
2:28here on the left hand side. I've got my
2:31login account down below. I've got also
2:34my projects and uh for each projects
2:37I've got chats like a list of chats etc.
2:41When you click on the dotted icon here
2:43you have option to you have the option
2:45to share the project to rename it
2:47project to to see the homepage of the
2:50project
2:52all the things that you can do here. So
2:54I pretty much copied or at least
2:57personalized my own interface to work
3:00the best. Now this is not a finalized
3:03interface necessarily but it's one of
3:06iterations one of the many iteration I
3:08did in terms of uh seeing comparing it
3:11to Quen Studio. You'll see here on the
3:13left hand side is very similar again you
3:15have the option to do new charts to
3:17search for chats. You also have the
3:20options to see the libraries with all of
3:22the media etc. And uh you can organize
3:26your chats into folders or projects. My
3:29interface here does pretty much the same
3:31things. Other thing you can do, you can
3:33personalize your the voice that you can
3:37use to answer your back. So, the AI can
3:40have a custom voice, and I'm going to
3:42show you how that works in a second. You
3:44have also a narrator voice that you can
3:46customize. You can change the character
3:50personality as well. If you want to, you
3:52can do that. and also add other
3:55graphically appealing widgets in the
3:58interface. In terms of saving your
4:01charts or exporting your charts, I made
4:03it in such a way that you can export it
4:05retaining all the media that you use in
4:07that specific thread. Exporting the same
4:10thing and then when you click on next
4:12launching and everything launching
4:15everything you have the uh instructions
4:17on how to use it. Now this chat this
4:20local AI chat works with LM studio in
4:23the background which works as a brain
4:26and with all talks uh TTS this is the
4:30one that I use so far and then I tested
4:33I tested some others which I wasn't
4:36really happy about uh like such as
4:38Google Microsoft Azour then I also
4:42tested um Piper and others as well and
4:46the voices were horrible.
4:48at least in my opinion. So, all is the
4:51one that I chose at the end. Of course,
4:54I've got omnivoice in confi, but there
4:57was no easy way to interface to link
5:00confui with an interface like this or at
5:04least I wasn't able to to actually do
5:06it. So, I prefer to have this
5:10configuration instead. So all you need
5:12to do to make this work is to have LM
5:15Studio installed with a model downloaded
5:19in it and then you need to have the
5:22launcher which is a file that I created
5:25which will install alls TTS into your
5:28machine and will also launch the ST
5:31server as well. Now the old talk to
5:33talks was a software was developed up to
5:38December 2025
5:40and after that uh it wasn't I don't
5:43think it was even touched by the
5:45developers I think for what I saw but it
5:49is a bit buggy when you install it if
5:51you go to the old talks DTS on a face or
5:54yeah I think it's on a face or github
5:57you might find some uh issues when you
6:00install it I think It's because of
6:03Python version. I had some several
6:07issues installing this and I found a
6:09solution by installing the right Python
6:13version which was used when they develop
6:16all talks DTS and that worked.
6:19So I also discovered a bug through an
6:21LLM. I'm not a genius. I use NLM. I use
6:24Quen Studio for that. And by doing it
6:27back and forth, I eventually discovered
6:30that the, you know, it discovered there
6:32was a glitch on the original code. And
6:36we fixed it. He fixed it for me. And now
6:39it works fine. And um then again, this
6:44launcher is going to be just a double
6:46click for you. So you all you need to do
6:47is simply double click on it and it will
6:49launch. I'm going to show you exactly
6:50how I do it. And also you need to just
6:54open the interface on your browser or
6:56any browser you prefer and you're ready
7:00to go. So you can of course customize
7:04your own voice and easily upload it or
7:08put it into the folder that I'm going to
7:09show you in a second and you can back up
7:11your conversations etc. If you change
7:14your mind, you want to use some other
7:16interface. Then I'm going to click on
7:17start chatting and this is how the
7:20interface looks like. It looks quite
7:21plain but you can customize quite a few
7:23things. So the first things we're going
7:25to do here on the left hand side is of
7:29course creating new chat if you want to.
7:31Of course on top you have the logo here
7:33if you want to find out about the
7:34instructions. You can just click on it
7:36and it will it's going to open this
7:38window again the instruction window and
7:40the features list. Then you can click on
7:42close if you want to. Then you can go to
7:44new chat. This will uh will open a new
7:46chart down here which you can rename by
7:49just hovering the cursor over the three
7:51dotted icon here. You can rename, you
7:53can clone the chart if you want to. You
7:55can export it as a HTML. You can move it
7:58into a project or into a folder in this
8:01case or you can delete.
8:04And you can also search for chats by
8:06keywords and it will automatically
8:08filter out this list down here. And you
8:11can choose a media folder. This will
8:14allow you to save all the uploaded
8:16images that you use during the chat if
8:19you want to. If you use a NLM model who
8:22has uh that has a vision in it, you can
8:25basically uh gather them all together in
8:28one place. So all I'm going to do, I'm
8:29going to click on media and I'm going to
8:31choose my folder. This case here is
8:33under the desktop under the Aura AI
8:36folder and it's called media chat. But
8:39you can choose any folder you like and
8:41click on select folder. And now it's
8:43going to ask me if I want to allow the
8:45site to uh to edit it. I'm going to
8:47click on allow. It says here media
8:50folder linked media chart. So that is a
8:52confirmation that it works. Now the
8:55other thing we need to make sure is the
8:58connection. At the moment the connection
9:00here has three LED. They are all green
9:03means that the chat is well connected
9:07with LM Studio and all talks. But I'm
9:09going to close this and I'm going to
9:11show you this from scratch how it works.
9:13So, first things we're going to do,
9:14we're going to close this and we're
9:16going to actually close the entire
9:19browser and we're going to also close
9:21the the PowerShell and LM Studio as
9:24well. I'm going to just start from
9:25scratch completely. And so when you
9:29install this uh folder, I'm going to uh
9:32you're going to find the zip file. After
9:33you unzip it, you're going to see a file
9:36called start local voice chat or
9:41something similar. But I will basically
9:42give you the step-by-step instructions
9:44when I create the the guide. But you see
9:48you got start local voice chat. This
9:50will when you double click on it, it
9:51will open two powershell. One will
9:54launch the old talk TTS and the other
9:56one will launch will launch the driver
9:59for the speech to talk to allow you to
10:02talk through the mic and dictate what
10:04you want. So you don't need to do
10:06anything else. Once you double click on
10:08it, you just minimize these two windows
10:10and forget about them. The second thing
10:12you need to do is simply open LM Studio.
10:15So, open LM Studio if you don't have it
10:18downloaded and then go into the to the
10:21search and just search a model that you
10:25want to chat with. At the moment, I've
10:27got already a few models in my list. And
10:31I'm going to choose a couple. One is a
10:34small model so we can have a quick
10:37response and have the feel that we
10:39chatting with our Jarvis AI. And this is
10:43going to be the the 12 GB model which
10:46has also a vision. And also we have the
10:49GPT OSS which is 22 GB. And we have also
10:54the Gemma 412B obliterated which is 12.7
10:59GB which is another small LLM that fits
11:02in my graphic card. By the way, I'm
11:04using an RTX 5070 Ti with 16 GB of RAM.
11:08I also have 96 GB of system RAM for
11:12those of you who are wondering what I've
11:13got and a Ryzen 7 7700 3.8 GHz as CPU.
11:19So that's what I've got. But in order to
11:21make Aura AI chat with you and have a
11:24pretty responsive and quick reply is
11:28good to have a model that fits into your
11:30graphic card and does not offload
11:33anything on your system RAM. So, I'm
11:35going to go into this tab here on the
11:37left hand side, which is developer. And
11:39here, I'm going to load that model. It's
11:41already loaded, but I'm going to load my
11:43model from here. All you need to do is
11:44simply go to load model up here. I'm
11:47going to choose something small. I'm
11:48going to choose this one here, 11 GB.
11:50It's going to load model fairly quickly.
11:53The other thing you need to make sure is
11:55under server settings, you make sure
11:58that enable cores is enabled. So, make
12:01sure that one is on. And that basically
12:03allows you to to share LM Studio with
12:06third party software. In this case, our
12:08TTS and SDT.
12:11Then after we done that, we're going to
12:13go to status
12:15here and running. Make sure that this is
12:17green. And that's it. That's all you
12:20need to do here in LM Studio. Once you
12:23launch this start uh local voice chat
12:27and LM studio and you load your model,
12:30now we are ready to open our Aura AI and
12:34chat with it. So I'm going to go into my
12:36desktop. I'm going to go to Aura AI and
12:39click on my HTML. And you can see now
12:43that the connection here are all green.
12:45Now for each of these elements if you
12:47don't know what they do you have a
12:49little question mark next to it and it
12:51will tell you what that lead is about
12:53LED is about or well the element is
12:55about in this case the first here is LM
12:57studio and the second one is our old
13:00talks TTS and the third one is our
13:03microphone ear so is ST server which
13:07allows us to actually use the mic. So
13:09now we're going to test the the AI and
13:11the mic if it works. So I'm going to go
13:13into the mic down here. But before I do
13:15that, I've got a couple of check boxes
13:17here. So the first one here is called
13:19stream. If you don't know what that
13:20does, make sure you hover the cursor
13:22over the the question mark and it tells
13:25you when on the AI replies appears word
13:29by word like a like a conversation. When
13:31off, the chat waits until the full
13:34answer is ready before showing anything.
13:37I wanted to have that feature. So while
13:39I'm talking it type something just to
13:41have some sort of a satisfaction of see
13:43if it's responsive or if actually is
13:46listening properly of what I'm saying
13:49and also a way to test your mic levels
13:51as well. The other one here is a is the
13:55auto send response. So if I want after I
13:58talk I can click on the mic again and
14:00that will auto send the text into the AI
14:05conversation. If I don't want to do
14:07that, I can uncheck that box and still
14:10see if everything has been typed
14:12correctly before I send it to the AI.
14:15I'm going to click on the mic and I'm
14:16going to start to talk. And of course,
14:18every time you uh click on the mic, you
14:20have to agree or to allow the browser to
14:24access your mic. And this will happen
14:26each time you use the mic because it's a
14:29local HTML file. And I think all the
14:33browsers do this unless you have a
14:35browser that automatically accept that.
14:37But I think after you allow this allow
14:40while busying the site, it won't bother
14:42you anymore once you retype the
14:44microphone. But each time you relaunch
14:46Aura AI, you would have to allow it at
14:48least once. I'm going to click on the
14:50first one. Hi, how are you? I am testing
14:55this local AI and wanted to just check
15:00if what I'm saying is been typed
15:03correctly. Let me know what you think
15:06about this interface. Now I clicked on
15:09the mic and now everything is typed
15:12here. Probably one thing I have to fix
15:14here. I wanted to be able to see the
15:16whole paragraph because it looks like it
15:18does this all in one line. But this is
15:20something I'm going to fix probably
15:21later on. So, it looks like it's typed
15:24everything correctly. If I'm happy with
15:25that, I'm going to click on send. We
15:27also have an option to add a an emoji or
15:31just to make a little bit the
15:32conversation a little bit more fun and
15:34click on send. So now the LM Studio that
15:40works in the background is loading the
15:43my reply my text and you can see how
15:46many tokens is taking. This is a small a
15:50relatively small model 12.67 GB. It
15:53shouldn't really bother my RTX, but we
15:56also recording. So now it's basically
15:59cooking that in the background. It's
16:01going to now type the response. And I'm
16:04going to hear also the response as well.
16:07And I'm going to show you also how to
16:09customize the voices here on the right
16:12hand side of the interface.
16:15The first time you launch this, it might
16:17take a little bit longer, but as you can
16:19see, after that, it should be fairly
16:21quick. As you can see, now it's typing
16:23everything. And it's not going to type
16:25everything all in one chunk. It just
16:27does this uh gradually.
16:30And it says, "Hello, I'm doing well.
16:32Thank you for asking to answer your
16:34questions." And it's going to basically
16:35answer. So, this is a quite long
16:38response. And you can also customize the
16:42response here on the right hand side
16:44under reply length. This will basically
16:47give you the option to go all the way up
16:51to 8192. I think if you go and if you
16:54want to find out how to use all these uh
16:57options, just need to go on the question
17:00mark and it will tell you range from 64
17:02to 81 192. Now it's about to talk. So
17:05I'm going to shut uh shut my mouth and
17:07let it talk. But as you can see here,
17:10you have all the explanation how to use
17:11it.
17:12>> Hello, I am doing well. Thank you for
17:15asking. To answer your questions, one,
17:18is what you typed correct? Yes,
17:22everything you typed is perfectly
17:24correct and readable. Your sentence
17:26structure is clear, the grammar is
17:28accurate, and there are no typos in your
17:31message. Hi, how are you? I am
17:34>> Okay, I'm gonna stop there because
17:35otherwise gonna read the whole thing.
17:37Um, and by the way, you can also choose
17:40to stop the speech if you decide to. You
17:43can also do a bunch of things here at
17:46the bottom. You can read aloud if you
17:48want that to reread. You can edit the
17:50response. You can regenerate. You can
17:52branch in new chats. And you can delete
17:54as well. And you can do the same thing
17:56on your reply on your text. When you
17:59hover over the uh, you know, on the
18:01message, you see the option to edit and
18:03to delete. You can also change the image
18:06avatar for you and your AI bot and I'm
18:10going to show you how to do that in a
18:11second. But for the voice, you have here
18:13a list of voices here on the right hand
18:15side. You have female Z01 Wave. So when
18:20you click on it, you see you have a list
18:22of all available voices. You can also
18:25add your own voices. And by the way, if
18:27you want to add your own voice, all you
18:29need to do is simply go to the folder,
18:31the old talks TTS folder, which you're
18:35going to have after you install it.
18:36Under voices, you can simply drag and
18:39drop all the W file that you like. Of
18:41course, don't import something that is 1
18:45hour long. Trying to make it that it's
18:47either 10 seconds, 30 seconds, one
18:50minute maximum. I think two minutes or
18:52something like that. I had that. I added
18:55some of them quite long and it works
18:58fine so far. So, make sure you add them
19:00here. After you add them here, you need
19:03to just refresh the page by clicking on
19:06the refresh button and they will appear
19:08on this list. If I want to select any
19:11other any other voice, I can just go
19:13here and I can go for instance to let's
19:17say male01 and I can test the voice by
19:20clicking on test voice here at the
19:21bottom. Hello. This is a test of your
19:23local text speech voice. It works.
19:27>> Okay. So, if I want to choose a
19:28different one, I can go to let's say
19:30female 01.
19:32>> Hello. This is a test of your local text
19:35to speech voice. It works.
19:37>> So, I quite like this voices. They're
19:40not bad. And the fact that I can
19:42customize mine and I can put mine in,
19:44it's quite nice. The very reason why I
19:46created this uh local chat interface is
19:49because I hated using Silly Tavern. I
19:51hated using those uh interface. I have
19:54to open text generation web UI and then
19:56I have to open silly tavern and then it
19:58they weren't linking properly the web
20:00bugs. So I decided I'm going to do my
20:03own local AI. And of course you need to
20:07launch at least LM Studio and you need
20:09to double click on that uh file that
20:11will launch both servers. But after you
20:15done that, it's the easiest way. You
20:16double click there, you open LM Studio
20:18and then you open this and it's ready to
20:20go. You see the connection already going
20:23and make I made it as easy as possible
20:25for you to use. In terms of linking LM
20:29Studio with this uh just I want to
20:32reiterate that you'll find LM Studio
20:34here on top which is already have the
20:37HTTP uh address which is this one here.
20:40Make sure at the end you have a forward
20:42slashv1 which you probably going to find
20:44already in the installation. And on the
20:47lm studio you also see here the uh it
20:50says is reachable at 18 1271234
20:55but it doesn't mention slashv1. So that
20:58it's quite important to actually add
21:00that into your interface in order for
21:02them to talk. The other thing you can do
21:04you can customize your character. You
21:05can add here a personality. Let's say I
21:08would say something like uh be brief in
21:12your answers and to the point. So I can
21:18do that. Um I can also select and
21:22narrate. If I want to have a narration,
21:24let's say if they answer an expl and you
21:26have like an explanation like a scene
21:28narration and you want to have a
21:30different voice, you can just simply
21:33tick this box and choose the voice you
21:35want to have as narrator. And the voice
21:38on top would be the voice of your
21:40Jarvis, your local AI assistant. In
21:44terms of uh other options here, you got
21:46down below is voice cap. Voice cap is is
21:50basically how many characters the voice
21:52will read after generate a response. So
21:56I put it zero. So it means it will read
21:58the whole paragraph. But if you want to
21:59only have for instance the first
22:01paragraph or so all you need to do is
22:03simply choose any many any characters
22:06from here and you can just do that
22:08creativity. So all these options here I
22:11saw some of them in silly tabvern which
22:13has like hundreds of those and have some
22:17crazy terminology which I don't know I
22:19need a translator to understand that
22:21interface. So I thought let me just make
22:24one which is super clear for everyone
22:26and easy to use. So the creativity
22:30allows you to to range from zero to two.
22:33So when is zero is dry, factual and
22:37predictable. When is for instance 07 is
22:39balanced as a default is 07. When is 1.5
22:43and up is going to be wild, poetic,
22:45imaginative, etc. So I put 1.8. This is
22:48going to be more like a test to see how
22:51it behaves. In terms of reply length, I
22:54put 20 48, but you can choose a range
22:57that goes from 64 all the way to 81 92.
23:02So this is something you can have a very
23:05thorough answer if you want to or a very
23:07short answer by just limiting this. The
23:10variety will let you choose between zero
23:13to one. And this is going to be how
23:16adventurous the AI is with a word
23:19choice. So if you choose something like
23:2209 is going to be if you want the the
23:26the chat to be balanced. I put one
23:29because I want the AI to be varied when
23:32it talks. The repetition also same
23:34thing. If you have a range from zero to
23:37two, how hard you want the AI to try not
23:41to repeat itself. I put 1.1. And in
23:45terms of memory, this will uh I would
23:47probably suggest you to leave this at
23:48zero. But this will remember all the
23:51conversation within the same thread. But
23:54if you want that to be less to remember
23:56only the last 10 or 15 replies, you can
23:59just choose 10 or 15 here on this
24:01number. In terms of look, and this is
24:03the fun part, you can choose your own
24:06avatar. In this case, I put my own uh
24:08image here. If I want to change my
24:10avatar, I can just go to my avatar and
24:13choose a different image. So, if I go to
24:17let's say downloads, I'm going to go to
24:20image. I'm going to choose an image of
24:23me. Let's say this one here. And this
24:25will change the image avatar for me. If
24:28I want to have my AI avatar as well, I
24:31can click on AI avatar and I can choose
24:34an image. Let's say on a desktop. Let's
24:36say I've got this image here. And by the
24:38way, you can choose either images with
24:40with or without the background. And I
24:42show you why would be beneficial to have
24:45that. So I'm going to choose let's say
24:47this one here with background and click
24:50on open. This will open the the image
24:53avatar here on the uh left hand side as
24:56well as here on the circle.
24:58Now if I want to customize this, if I
25:00want to move it around, I can just go
25:02down here and uh I can basically change
25:06the portrait size. So this will change
25:08the portrait size and as you can see
25:10goes beyond my chat. I can move it
25:13around. So I can click on uh move mode
25:16and this will allow me to move this
25:18image wherever I want in the chat. Let's
25:20say I want to have it here and I want to
25:23have it a little bit smaller. And you
25:26also have the option to flip it. If I
25:27click on flip, it will flip it as well.
25:30And this is purely graphic. If you want
25:32to have like a more interactive
25:34experience with your AI, you want to
25:36have it like this. You want to have
25:38whatever image you like. After you
25:40finish, you can click on move mode off
25:42and the portrait and the fade. This will
25:46allow you to fade your charts in case
25:48you have a background. And I'm going to
25:50show you how to choose a background. I'm
25:52going I'm going I'm going to go to my
25:54background here and I'm going to go into
25:57pictures and choose one of these
25:58background here and click on open. And
26:01this will open the background here. Now
26:03I can have the charts to be completely
26:05solid here, solid bubbles, or I can have
26:08a little bit shaded. It's up to you how
26:11you want to have it. This is purely
26:13graphic. After you finish that, you can
26:16decide to takeick it off if you want to
26:18have it solid or have it faded. It's up
26:20to you. And you can also reset the
26:22portrait position. So if I click on
26:24that, it's going to reset the position
26:25of the portrait there. If you want to
26:27move it again, you tick the box, you
26:29move it around, and you tick the box
26:31again once you finish the to position
26:33it. So, another thing you can do is back
26:36up everything. This will back up all of
26:38your chats in a zip file and it will
26:41organize them into conversation. You can
26:44restore a backup by clicking on restore.
26:47Select the restore zip file and it will
26:50restore everything in your chat
26:52interface. and the test voice. We
26:54already saw it before how it works. So,
26:57as you can see, this is something that
27:00you can use for voice. I'm going to test
27:03again with another reply. And this is
27:06going to be more like just a short
27:09conversation. I'm going to tick this
27:11box. This will basically auto send my
27:15reply. And click on the microphone
27:17again. Hi, how are you today? So, I
27:20clicked on the red button so that it
27:24will automatically send it for me.
27:27>> I'm doing well. How can I help you with
27:29your testing today?
27:30>> I'm testing the mic and how the
27:33interface behaves with the text. Can you
27:37help me understand about a product that
27:41I just bought? So, I'm just, you know,
27:44improvising here just to see if he can
27:46answer me fairly quick. Again, I'm using
27:49an 11 GB model, so the response is
27:53fairly quick.
27:54>> Yes, I can certainly help with that.
27:57Please let me know the name of the
27:58product and any specific features or
28:00concepts you would like explain.
28:02>> I've got an Herman Miller chair that I
28:04just purchased secondhand, and I would
28:07like to know where I could find the
28:08cylinder replacement for it. So this uh
28:12specific model that I loaded, I think
28:15it's uh updated up to the December 2025.
28:18So it will probably come up with
28:20whatever knowledge he has about it. But
28:23as you can see, he's already replying
28:26and uh and basically it's already
28:27telling me it's asking me about the what
28:30model.
28:32Now, it depends on how long the response
28:35is, but if you want to have a back and
28:37forward response, you probably just want
28:39to have different kind of conversations.
28:41But for this kind of stuff, you can just
28:44uh let it run. To identify the correct
28:46gas lift cylinder replacement for your
28:48Herman Miller chair, follow these steps.
28:51One, identify the chair model. Herman
28:54Miller uses different cylinders
28:56depending on the model, eg Aeron, Mirror
29:002. You must know which model you have
29:02before purchasing a replacement. Two,
29:04where to buy Owen part's original
29:06manufacturer.
29:07>> Okay, I'm going to stop there, but as
29:09you can see, it's answering you. It's uh
29:12fairly quick. Now, I can also attach
29:14images. Now, as you can see here, I've
29:17got media. When I click on media, I'm
29:19going to select my uh media folder. At
29:23the moment, when I open the chat for the
29:25very first time, after let's say one day
29:28or two, I reopen it. I need to go and
29:30select the media because apparently it
29:32forgets it. But I will fix this uh
29:35probably later on. After you select your
29:37media now, if I go to upload an image,
29:41um I'm going to ask to describe the
29:42image. I can do that. So for instance,
29:45I'm going to go into LM Studio. At the
29:46moment, this model does not have vision.
29:49So I would have to make sure that the
29:51model I load here has a vision feature.
29:54I'm going to go for this one here, 20
29:57gigabyte. Okay. After I done that, I'm
30:00going to go here. I'm going to go to the
30:02attachment here and I'm going to select
30:04an image from the desktop. Let's say I'm
30:08going to choose an image of I'm going to
30:10choose this image.
30:12So, I attach the image. I'm going to
30:15dictate this. Please describe the image
30:18I attached. And as you can see now, the
30:21this interface is going to do the job.
30:23It's going to analyze the image and it's
30:26going to describe it for me. I wanted to
30:28show you here under the desktop under
30:32Aura AI under media chart. When I double
30:35click there, as you can see, the image
30:36has been saved here as well. I've done
30:38it this before. I was testing it. So
30:41each time you upload an image in your
30:42thread, whatever charts you are in, it's
30:45going to be saved in that folder. I
30:46found it useful to have that um attached
30:50and well organized in the same chat. If
30:53you have any idea of any features that I
30:55haven't thought about, please leave it
30:57in the comment down below. It would be
30:58very useful to uh to add those additions
31:02if uh if I found it useful. If you find
31:04it useful, I would uh try to implement
31:07them into future versions of this
31:10interface. So another thing you can do
31:12here in the interface uh is to save the
31:15character. So if you are changing the
31:17personality, if you uh have let's say
31:20done you have done all these changes
31:22here and you want to save it, all you
31:23need to do is simply click on save
31:25character, give it a name. I'm going to
31:26call it let's say I don't know Jarvis or
31:30whatever and click okay. I'm going to
31:32have that saved into this list and I can
31:36choose this next time I open the
31:38interface or if I want to uh delete it,
31:40I can click on delete and this will
31:42delete the character for me. So that is
31:46all I've done so far with this. I spend
31:49a couple of days at the moment and 27 or
31:51more iterations to you know to to write
31:54this HTML. Of course I didn't write it.
31:58The LLM did it but I basically test it
32:00and amend it so far. If you want to know
32:03how I did this, please click the link
32:06down below which I basically share all
32:08my um thought process and all the what I
32:11went through in order to develop this.
32:14So you see exactly behind the scene on
32:16how I did it if you are curious to know.
32:19Basically you're also going to get the
32:21interface itself with that package. It's
32:24just a small little uh contribution to
32:27support the channel and hopefully you're
32:29going to use this chat as well. So let's
32:32have a look what's going on here with
32:33the with the model. Bear in mind that
32:36when you open the model for for the very
32:38first time it's going to take a little
32:39bit longer to answer your queries. But
32:42as you can see, it's almost done here.
32:45And but after it loads the uh the model,
32:49the subsequent replies should be fairly
32:52quick or reasonably quick. At the
32:54moment, I chosen a 20 GB model which now
32:58goes over my graphic card which is 16
33:00GB. So some of them is offloaded into
33:02the memory system. Let's see if it's
33:05actually working and describing the
33:07image. And as you can see now is
33:09answering the question is describing the
33:12image. The portrait features an African
33:14woman's face in closeup against a warm
33:16cream background and etc. is doing the
33:19whole explanation and this is done you
33:23know word by word at the moment because
33:24I asked the you know the LLM to uh
33:27develop this HTML in such a way that
33:30when the LM studio is analyzing the
33:34answer it will write you know straight
33:36away without me waiting for very long
33:38time. So I can actually start to read
33:41the answer before it finishes
33:44completely.
33:46Hello everybody. In this video, I'm
33:47going to show you how to install all
33:50talks DTS into your machine and how to
33:52use my patcher that you find in the zip
33:55file. So, first things first, let's go
33:58into the desktop and we're going to go
34:00into Google and type all talks DTS on
34:03screen here. The first search should be
34:06the one that we're looking for, which is
34:08all talks DTS. Going to click on that.
34:11And here you'll see we are under GitHub.
34:14So, all you need to do is simply go into
34:16the code here on top and download the
34:19zip file. So, after you download the zip
34:22file, you're going to make this a little
34:24bit smaller.
34:27The zip file should be on your
34:29downloads, which is this one here. We're
34:31going to extract that. Uh, let me just
34:33minimize this for a second. We're going
34:35to go into right click, extract all, and
34:39we're going to extract it here on the
34:42download folder. Now, we're going to
34:44double click that. And this is the
34:45folder that we need to drag into your
34:48into your AI folder or into your system.
34:50I'm going to drag it into my AI folder.
34:55Move it there. So, it's here. I'm going
34:57to click on it. I'm going to name it
35:00test_all
35:04talk. This is just for you guys that you
35:07can see it. Now, when I double click
35:09here, now you have all these files. We
35:12need to basically start the setup here.
35:14We're going to double click on at
35:16setup.bat.
35:17I'm going to double click there. I'm
35:19going to run it.
35:21And now the choice here are two. We're
35:24going to use it as a standalone
35:26application. We're going to press two
35:28and press return. And then we're going
35:30to press one here to install the the
35:33files into our system. This is going to
35:35take around 5 to 10 minutes depending of
35:38your internet speed. and it's going to
35:41install all the files needed for all
35:44talks to work and also the environment
35:47as well. Be patient and uh let's wait
35:50until this is done.
35:55Now, if it the installation failed or
35:57failed the the AV to run properly here
36:01to build a wheel for AV, do not worry
36:04because my patch is going to do the job
36:06and uh fix these issues. So I spend uh
36:09two or three days only to sort this out.
36:12So as you can see now it's ready to be
36:15launched. What we need to do is simply
36:16just press any key to close the
36:21PowerShell and then we're going to press
36:23nine here to escape and go back here.
36:26Now we need to use our patcher. When we
36:28use our patcher, we're gonna go into the
36:31zip file that you find in the download.
36:35And
36:37we're going to go into our desktop.
36:39That's what my Aura installer is. And
36:42there we go. So, these are the file you
36:44find into the zip file I gave you. And
36:47uh the one that we need to look at is
36:50Aura AI_installer.bot.
36:53So all you need to do is simply go back
36:55into the main folder called test all
36:58talk or whatever name you gave it and
37:01basically we drag this folder and drop
37:03it into aura installer.bat.
37:06This will start to install
37:10all the files that that are needed for
37:12uh text to speech. Actually, it's going
37:14to fix texttospech files from the
37:17original installation and it's going to
37:20install ST the speechtoalk server as
37:24well. So, everything is going to run
37:26hopefully smoothly here. And after this
37:30is finished, you're going to see a
37:34launcher popping up on your desktop and
37:36there's going to be a file that you need
37:38to double click on. So you are ready to
37:40go and use the local AI directly. Let's
37:44wait until this one is done and then we
37:46are going to be back. The installation
37:48has finished. Now it says here, press
37:50any key to continue. I'm going to press
37:52return to close it. And on the desktop
37:55now we have start aura AI.but. But so
37:59before we double click on that, we need
38:01to open LM Studio and of course load our
38:04model which is our brain. So we're going
38:06to go into LM Studio here. So make sure
38:09you download LM Studio. The interface is
38:12going to look like this. And uh all you
38:14need to do is simply go to the search
38:16here and download the model you want. I
38:19downloaded already a few models here.
38:21I'm going to my developer here on the
38:25left hand side on the top tab and I'm
38:28going to load my model. Going to eject
38:30this one because it's already loaded.
38:32All you need to do here, you're probably
38:34going to see this page. And then you
38:36need to go to load model here on the top
38:38right hand corner. I'm going to choose a
38:40small model so we can test it quickly.
38:42The Gemma 412B, the the first one here,
38:46which is going to be fairly quick to
38:47load. The other thing you need to make
38:49sure is that you go into server settings
38:53and make sure that enable cause is on.
38:56This enable enables uh LM studio to
39:00share the model with other servers in
39:04this case our ST and TTS server. So make
39:07sure that you turn that on and then once
39:09this is loaded you need to just the uh
39:12click on the status to run it. That's
39:15it. You don't have to do anything else
39:17here. Okay, just need to minimize this
39:18and then we are ready to double click on
39:20start aura.bat.
39:23We're going to double click there. This
39:26is going to open all talks TTS and our
39:29ear as well and ST. And this is how the
39:34interface looks like at the time of this
39:37recording. And this is how the interface
39:39going to look like. and make sure that
39:41this connection here on top are all
39:43green at the moment is still loading our
39:46shell here. The power shell is still
39:48loading. The ST the speech to talk is
39:52already loaded. You can simply um
39:55minimize that and all talks now will
39:58start and if you uh start all talks for
40:01the very first time is going to download
40:04the models. Next time you relaunch it uh
40:08you don't have to do this. is going to
40:09be much faster of course, but the very
40:11first time you have to wait until all
40:13these models are downloaded. Here you
40:16got 1.8 GB to go through and another
40:18couple as well. So, make sure you wait
40:21for that. And then on the interface,
40:24you're going to have all of the uh
40:26traffic lights here in green. As you can
40:28see, my ST server now is loaded in port
40:327852.
40:34And the TTS now is waiting until this
40:37one is finished. And now you be ready to
40:40go and use your interface on your
40:44machine locally, privately. So let's
40:47wait just a few seconds until this is
40:49finished and then we're going to be
40:51back. That's it. So now we need to just
40:53uh minimize this and uh let's check if
40:56our chart now is all green. If it's not
40:59green yet, do not worry. Just refresh
41:02here. this little blue button on the
41:04side. And if it hasn't done it yet,
41:08don't worry. It's still loading. So,
41:10let's go back into our PowerShell and
41:12see if it's still cooking something. It
41:14says now ready.
41:17And now we going to minimize this. And
41:18as you can see now, the traffic light is
41:20all green. And if you want to check the
41:23voices, if you don't see them here,
41:24click on the refresh button. And now we
41:27got all the voices that you can just
41:30click and test. For those of you who are
41:32interested to know on how I developed
41:34this interface from scratch and you want
41:36to watch over my shoulder on how I did
41:38it, sort of like behind the scene, watch
41:41my following uh videos here on YouTube.
41:44I'm going to be releasing episode one,
41:46two, and three. This is actually episode
41:484, the one you watched, and I showcased
41:51the interface. But in episode one, two,
41:54and three, you're going to see exactly
41:57how I developed this uh this app, this
42:00little HTML. All the nitty-gritty
42:03details that you wanted to see are shown
42:06in the following episode. So stay tuned,
42:08subscribe, and I'll see you very soon in
42:11the next video. Bye.