stable-diffusion / AI - PART ONE

[maxhitman]

Practice what you know, and it will help to make clear what now
you do not know.
- Rembrandt

Anyway I was going to write a massive tutorial but the closest one I found to be correct and easy as fuck to follow is the one below however......

At 4 minutes he forgets to say if you should use FP16 or BF16, there is hardly any difference so choose FP16 and hit enter.

http://youtube.com/watch?v=9MT1n97ITaE

Now I did this, instead of following his way to install Lora I just went to the Lora page where he says to go which is here,

http://github.com/bmaltais/kohya_ss

And what I did was to click the GREEN square at the top right where the word code is and downloaded the Zip to My Desktop and extracted it then clicked on Upgrade file as described and after that I just copied the kohya_ss folder to My Documents as described in the video.

So basically I downloaded the Zip folder instead of running scripts to install with Git, but I did install Visual Studio and run Set-ExecutionPolicy Unrestricted as decribed in the video.

That means I have a clean folder that is on the Desktop of Kohya that I can keep on a spare drive if I mess the folder up that is located in My Documents.

Please read..............

Where do you put Lora files you might ask....so if you go to the website Civitai and download a Lora file it needs to go in the Lora folder but not in Kohya it goes in stable-diffusion in the Lora folder.

For example - F:\stable-diffusion\stable-diffusion-webui\models\Lora

And as the video above explains you must restart stable-diffusion and the Command Prompt and then click on show extra networks (1st) then on Lora (2nd) and Lora files will appear, dont for get to insert the Lora shortcut in the positive prompt.



Lora shortcut




So if I created a Lora of Mia Goth or downloaded one and gave the file the name of miagoth18966 my positive prompt would look like this.

photograph of <lora:miagoth189661:1>, cleaning out the shitter with a tootbrush etc etc.....

[maxhitman]

I don’t paint dreams or nightmares, I paint my own reality.
- Frida Kahlo

[MaxLord]

Originally Posted by spreadwell

Just an observation, the jaw line is slightly off, chin is a bit mis-shapen, skin tone too bright and clear, eye colour is off a tad, hair too orange, and the conk needs a pinch, lol.

The jawline has been one of the trickiest parts honestly but that last batch, for some reason, seemed to cross the threshold into getting much closer to ideal.

[maxhitman]

Just one small positive thought in the morning can change your whole day.

[MaxLord]

I think after this bit I'm going to one last round of ten batches of six:



One thing that seems to be a running glitch I suppose is every so often I get half a batch of blue or green eyes despite putting in "brown eyes" as prompt. I don't want to overwhelm with too many parenthesis but I almost feel it's necessary for that detail.

[pelham456]

those coronation pix are something else! even the best of the images look "cartoonish" to me -- like a norman rockwell/saturday evening post cover. i don't see ANYTHING in any of those even close to the level that u guys here are working at. perhaps mr lloyd-jones needs to join PS?

any opinions here about midjourney vs stable-diffusion? or the others (DALL-E, etc)? is S/D considered the best; is that why u guys are focused on it?

it is surprising to me that MJ struggles so much with the UNION JACK. stripes going every which way in some of those pix!

i also rly love the carriage with the PERPENDICULAR WHEEL. great for tight corners, not so great GOING FORWARD!

Originally Posted by pelham456

those coronation pix are something else! even the best of the images look "cartoonish" to me -- like a norman rockwell/saturday evening post cover. i don't see ANYTHING in any of those even close to the level that u guys here are working at. perhaps mr lloyd-jones needs to join PS?

any opinions here about midjourney vs stable-diffusion? or the others (DALL-E, etc)? is S/D considered the best; is that why u guys are focused on it?

it is surprising to me that MJ struggles so much with the UNION JACK. stripes going every which way in some of those pix!

i also rly love the carriage with the PERPENDICULAR WHEEL. great for tight corners, not so great GOING FORWARD!

I would go stable-diffusion because you can install it without signing up to anything and because it is on your PC as supposed to a web site the images that you make stay on your PC and don't get shared somewhere if you don't want to share.

SD is better I think because the interface is easier to understand and it seems more people are using it so more resources and help.

Originally Posted by MaxLord

I think after this bit I'm going to one last round of ten batches of six:



One thing that seems to be a running glitch I suppose is every so often I get half a batch of blue or green eyes despite putting in "brown eyes" as prompt. I don't want to overwhelm with too many parenthesis but I almost feel it's necessary for that detail.

Much better on the jaw line and chin dimple and the nose is better, some people say that (brown eyes:1.2) is better than ((brown eyes))

I have also noticed a few prompts with this....

RAW, analogue style, 8k DSLR, Fujifilm XT3

You have to be careful as this can confuse the prompt and I shall explain why.


RAw, this is only used by DSLR so a digital format

Analogue is a film format, 35mm for example

8k that's also digital

DSLR being used with Analogue can't work

Fujifilm XT3 is also mirrorless so can't also be DSLR or 55mm

I would use a prompt like, DSLR, Canon 400D, Lens: EF-S18-55mm f/3.5-5.6 II, Raw, ISO 200, DOF

Or you will be sending the AI mixed messages