Trying to setup AI on my PC
richardandtracy
Posts: 7,685
I'm not sure if this is best in The Commons or in Technical Help, but I would appreciate some help.
I have been seeing some interesting picture to picture AI enhancements and am beginning to think it is something I need to know more about. So, maybe someone might be able to help me.
My situation is this: I have a Win 11 PC/3060 card, 64Gb RAM, Intel i5-13600k processor. Where it's slightly crippled is that it has no internet connection - no wifi (deadspot) and no cable (too far from router) and as a result, internet is not an option for me. So.. I need a standalone version. I am also a total cheapskate and would prefer to do my initial experimentation for nothing (if it's a dud, I don't want to waste money) and maybe shell out some money when I have a better idea of my limitations, the AI's limitations and any alternatives that are around. At work I have access to CopilotAI, and I have tried to get it to do a picture-picture improvement on the attached image. It was rejected because 1) too much bare flesh and 2) It won't work on photos of real people. What? That not a photo of a real person, you'd never recognise them even if it were due to the hair & sunglasses, and the dress is not seriously revealing. I suspect it must be programmed to be more uptight than a Victorian Maiden Aunt. So, my feeling is that I'd prefer fewer built in restrictions rather than more, some are OK if necessary.
Knowing very little about it, I thought Stable Diffusion might be suitable. I have transferred over Microsoft's Python and the Webui Stable Diffusion from GitHub (that was hard - I still have no idea how anyone understands GitHub) to a stick & ran from there. The Python installation went OK. However, Stable Diffusion seems to want to update itself online - and it can't. So, I'm left in a quandry.
Can anyone suggest a picture-picture AI system that's available as a complete, self-contained, package (zip installation preferred, like with DS content - I can understand that 
) with no external internet connection required? And what else would I need? I have seen talk of Lora's. What are they, where/how does one get them? If Lora's affect the picture, the sort of thing I'm looking for is heading towards the photoreal end rather than the NPR end of the scale, with people in landscapes or just people (4 fingers and a thumb on each hand preferred!). I'm afraid I don't yet know the extent of my ignorance, so am probably asking the wrong questions - if you think I am, please tell me. What would be really good is a pointer to a series of written/printable steps to follow to get it installed (no internet at my PC, so a YouTube video is as impossible as all other internet streamed/update stuff).
Many thanks in advance,
Regards,
Richard
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2026 Daz Productions Inc. All Rights Reserved.
Comments
I have very limited understanding of AI, although I have 'played' with it. So far as I know the AI engine that you install then needs to download 'models' which can be huge ( as in several Gigabytes ), and as you say this is generally handled by the engine making an internet connection and downloading what it needs. Not sure if a completely 'internet- not-required' AI installation is available, if there is one it will be a huge download.
That may be possible if I could know which ones are needed. There is an internet connection in the house and I can download things, it's just that the only practical place for my PC happens to be a wifi deadspot (4off 9" thick brick walls in the way) and on the far side of the property from the router (hence no cable). I 'just' need to download stuff onto a stick and then feed it into the PC from there.
Regards,
Richard
the only way you could download the standalone apps I use would be to use a different PC then copy them over
but things get written and verified so not sure how you would go
a Dlink or other WiFi USB dongle isn't that dear
you could even use your phone to hot spot if you have enough data
or buy a prepaid sim card with lots of data
I actually was on mobile internet in 2009 when I first used iClone, then Carrara DAZ and Poser,
first a dongle with a sim card then my Samsung Galaxy tablet as a hotspot
used it for years because my landline wasn't connected to the internet, then naked ADSL eventually came here
now I am on 5G, so on mobile again
as an aside
you know those 40K posts I have...
35K of them were reached 10 years ago because I was continuously chatting to about 5 people all day on my tablet in a handful of threads!
I was on it all day, on social media too
I actually stopped using a phone or tablet outside of my house, am the only person without a phone on me now
If it's any consolation I think I could sum up my experience with AI in two words, 'Frustrating' and 'Disappointing'.
Frustrating because it seems to be almost impossible to make it produce images that I would have thought were fairly straightforward. For example I was doing a render of two people having a conversation in a car, so I wanted a background image of houses, such as you might see if you looked out of the side window of a car. Despite framing my prompt in a hundred different ways the AI insited on giving me views down a road, or views of distand houses, or views that were looking from a low flying aircraft. I wasted a whole day on it, and I think I got two useable images. AI seems to insist on there being some central focus. If I ask for a view of trees in a jungle, I get an image of a path through the jungle, which kind of ruins the premise that this is unexplored territory really... just frustrating.
Disappointing because any attempt i made to feed in an image, and have the AI modify it resulted in an unusable mess. I had a photo of someone walking down a street, and i carefully erased all trace of the street and provided a prompt to put the person in a different realistic environment. I would have done better to have just overlayed the cutout of the person onto a background. I recall I asked AI to create an image of someone searching for shells on a beach, and the shells were bigger than the person, so apart from the unfeasable size error, not much searching would have been required, more like "There's one, and there, and that's a giant one and there's two merged together, and we're still fifty feet away".. I think, since I mentioned shells the AI felt obliged to make them front and centre.
In the end I use it for background fillers, such as posters.
I have been using the Comfyui setup/interface for quite some time now. You can do a lot with it, but it requires learning new ways. Coming from 3D DAZ/Poser, we are used to certain ways and some of that you need to unlearn. I use it mostly in image to image mode and also to animate images. It takes time to get good results. @background experience is pretty common , I think. It takes significant time to get good results. Prompting is actually quite hard if you want decent results beyond very basic stuff.
My suggestion is to start with a good introduction to setup video (not to follow along, but to get a feel for what is needed). I found this one early on and it helped me getting started:
https://www.youtube.com/watch?v=6dXpgL1-YdM
One word of caution: trying to run set this up on a computer without internet connection is going to be VERY frustrating. You will need to,constantly download, update stuff in the beginning. Once everything is running you don't need a connection, but during the setup phase it's really crucial.
ciao
TD
edit to add: I just noticed that the link is to a newer version of the video I watched, but it seems to be still quite good.
Our whole house is in a 4G weak spot (steep valley, mast blanketed by hills, at absolute best we have one signal bar), and when it rains, we must use wifi calling inside the house because there is zero 4G signal. There is never any 5G at home, never istalled in the local mast, and 3G shutdown was completed last year. 2G calls & text remain until 2033, also weak due to the mast being blanketed by hills, but 2G's not internet compatible. In the rain, the only way to get a 4G connection is to go halfway up the 30ft high garden bank to get a signal. So, I suspect a Wifi dongle won't work.I've not tried it, I must admit.
As for a wifi booster/repeater etc. I have tried one that was supposed to send the signal through the mains power cable, but with zero success. Not sure why, maybe the mains signal is very dirty - our house has power tapped off 2 of the 3 individual phase cables strung up on poles down the road. It's been like this since the village was connected to the grid in the 1930's and some (half or more) of our wiring dates back to the 1960's. Fortunately, none goes bact to the 1930's. Anyway, whatever the cause, the upshot is that the booster didn't work and I took it back.
Loading the standalone apps onto a connected PC and then copying the installation across. This might be theoretically possible. My wife has a 10yo laptop, which actually seems less powerful than my Lenovo tablet, but she's very protective of it. The chances of me being permitted to install AI on it are low, and then, for the rest of my life and all future PC's she ever has, I will be blamed for every fault, which will inevitably be caused by installing the AI. Not sure it's worth the risk...
Hmm. beginning to think the hassle of setting up AI on my PC is really not worth it in my current situation.
Regards,
Richard
Have you tried the Powerline WiFi extenders? I have been using the TP-Link extenders for a number of years. I have the main one in the upstairs room where the Router is. an extender in the kitchen next to the conservetory when I am out there and one in the shed at the bottom of the garden, I was intending moving the workstation out there but never did, still handy though if I am sitting in the garden :)
https://www.screwfix.com/p/tp-link-av1000-powerline-starter-kit/326KX?tc=SC9&gclsrc=aw.ds&gad_source=1&gad_campaignid=19823510362&gclid=Cj0KCQjw_IXQBhCkARIsADqELbKwKqyBqWoQUfX6lM6cV2TFkkV8aldqqBgHhwj_ZI0iVsZ1gMFuW-EaAtmrEALw_wcB ;
or work in the cloud on your phone
(sitting on a hill
)
they have ComfyUI now apparently (old video)
https://www.thinkdiffusion.com/?via=mdmz
link from video still works
I can use my own PC so not tried it
Actually, the one I tried looked very much like that. I got it through Curry's, but I do vaguely remember 'TP Link', not the model number, though. I remember it had two tiny buttons & 3 LED's on the base station one.
Regards,
Richard
I fear that, for me, your suggested option is the opposite of what I want, but may be the only option. I am not wildly keen on sitting on the hill next to our house playing with AI, when all the source images I may wish to try to 'improve' are at the bottom of the hill, locked away on an isolated PC. I could use a tablet in the house and the cloud stuff, but.. it's farming out something I'm used to doing on my equipment. Need to get my head round it.
On the top of the bank in our garden, no joke, but we can see the oval of the chimney pot opening. The garden slope is 1 in 3 (17 degrees) so is pretty steep, and tops out over 30ft up from the house. (Try mowing that once a year!)
Regards,
Richard
time for a man shed on the hill with a small tower and solar panels
I already have half a 30ft x 15ft shed at the bottom of the hill with a metal lathe, milling machine, Boxford 7" metal shaper, aluminium foundry, drill press, metal cutting bandsaw, 4 axis cnc router (X,Y,Z and rotation) and most hand guided power tools you can imagine... I'd be hard pressed to expand my dominion beyond adding a cnc metal lathe to the list.
Regards,
Richard
It wouldn't solve the downloading issue, but Krita has a Stable Diffusion plug-in which will, once installed, download the required models (it take a while even with a fairly good connection though). My attempts to use it, however, have pretty much matched background's, even when using regions according a video guide made specifically for Stable Diffusion in Krita.
Maybe you have a friend who has good Internet connectivity, and you could temporarily take your PC to their house, long enough to get evrything downloaded an running? You might need to make occasional further visits, if you decided you wanted more models, but at least you would have enough to try the thing out.
A brave man having an aluminium foundry, I think if i had one I'd be counting the days until the house burned down. Power tools I view with a healthy respect, although I have a lathe, a drill, and a small circular saw. As I said to my ex one time "Be careful with these, they won't even slow down if they touch flesh".. She did cut through the cable to the hedge trimmer one time, but the earth leakage breaker tripped so there was no electical hazard.
The internet issue is due to the location in my house, not the connection to the house.
So, to summarise: 1) move the PC or 2) forget the idea.
It's 2), then.
Many thanks for your help.
Regards,
Richard
The foundry is fun, actually. I have a very small one, melt up to 1kg of aluminium and pour it into green sand moulds. The fuel I use is pallet timber, and have got it hot enough to melt the concrete refractory lining the furnace. Only once has it gone quite wrong, then I had mixed too much water in the sand, and lots of steam was produced, turning the casting into foam. It's quite a feeling to take a pot of molten aluminium and cast it into something like a replacement finial for our cast iron fence out the front of the house. The attached image is a photo I took after breaking open the mould. The casting is in the raw, unfettled state, and isn't at all bad - even if I say so myself.
Regards,
Richard
My lady wife did that with the hedge trimmer too, four times in an afternoon and got quite stroppy with me over how long the repairs took me to do. Apparently I wasted a large part of her afternoon.
Regards,
Richard
You could try Fooocus. It has a good UI and once installed, doesn't need an internet connection to use. It is a great tool for inpainting AI into DS renders, use it often
Fooocus is actually what I use for images
but I thought the installer needed internet
afterwards it doesn't
it also is the only generattive AI app that runs on my ExFAT drive
Thanks, I'll take a look at fooocus then. And temporarily moving the PC.
Regards,
Richard
there has been a stand alone AI photo editing program popping up om my FB page. Don't remember what is was, sorry, but if I say it 3 times near my phone it should pop up pretty soon.
Keep trying. Knowing how to use AI can be useful too in the future...
I found both ComfyUI directly and via Pinokio provided good, "click to install" versions to experiment with. It would involve downloading the basic set-ups, then you might want to download some templates and workflows to play with until you get your sea legs. My biggest problem was being Mac-based and not having the patience to get all the workarounds worked around.
PInokio might be the quickest to set up and play with:
https://pinokio.co
I hope your 3060 is a 12gb model. Running local AI on low vram is no fun. You'll grind the heck of your HDD or SSD. One option to consider, assuming you can get basic internet to your computer, is Runpod. A 100gb persistent network drive is $10 per month. You download what AI models you want to use directly from the web to this drive. You use your browser to set up an instance when you want to work with AI, and use their GPUs to iterate. An RTX 5090 is $0.97 per hour. An RTX Pro 6000 with 96gb VRAM is $1.99 per hour. There are other options, but I stick with ComfyUi as it has plenty of default templates that are easy to use.
An initial iteration on the 6000 with image to video at 960 x 540 takes about two minutes. Follow up iterations take literally 10 seconds. Full 1920 x 1080 takes about 3 minutes. This is using 14b wan 2.2 models. For reference, my RTX 4070 12gb, using the same workflow and models takes 5 minutes to generate the half res sequence while beating the heck out of my NVME drive.
There are GGUF models that work with lower VRAM that might suit your needs. Do some YouTubeing on ComfyUI and you will see there are tons of options. It's a rabbit hole
Maybe cutting it four times was a subtle hint, that she thought you should be trimming the hedge
and she got stroppy because you didn't rise to the bait 
That's the way my Ex used to think.