AI image generation

Fama

Well-known member
I've been messing around with generating images locally (you basically need a decently beefy computer, and a modern gaming graphics card with preferably at least 8GB VRAM, although you might get by with 4GB) lately, since it's pretty fun and free. I'm using them for my pedal graphics, but I figured it might be fun to share some of the pictures that don't make the cut. If someone has a request I could maybe also do those. As far as I know, I own the copyright to the images I generate (or at least if anyone owns the copyright to them, it's me), so anyone can use what I post as they wish too. But I try to avoid "in the style of X", at least if they're still alive, just to sidestep any ethics discussions, which I don't really care about having.

I like to base the images on the names of the pedal, but I might also add additional descriptions or whatever, so I'll probably just post the name of the pedal:

Dragon Hound

ComfyUI_219542215804694_00001_.png ComfyUI_765177867195811_00002_.png
A couple using a "t-shirt graphic" style LORA (don't ask me what it is exactly, my understanding is that it's a sort of a "sub model" trained on a smaller image selection).

refiner_output_00029_.png refiner_output_00021_.png
refiner_output_00038_.png
"Who's a good little hellhound?", this is another LORA which does these yellow/red/black style images, pretty cool. I think I'll use parts of that last one in the actual pedal.

Also tried a couple of fantasy painting style things with the generic model, but nothing really caught my interest. Some of them came out quite cool, but I don't think they'd work that great on a pedal necessarily.
ComfyUI_temp_gsrep_00042_.png ComfyUI_temp_gsrep_00043_.png ComfyUI_temp_gsrep_00044_.png
 
Strawberry Red Overdrive

ComfyUI_temp_qaoot_00025_.png ComfyUI_temp_qaoot_00026_.png ComfyUI_temp_qaoot_00032_.png

"Follow the Strawberry Red Road", and in the third one we can see how the Strawberry Mother kicks all of the strawberries that are too weak to survive out of the nest. It can seem a bit cruel, but that is the way of the world.

Good Vibes

ComfyUI_temp_qaoot_00016_.png ComfyUI_temp_qaoot_00012_.png

These came out ok, but again not blown away by them, so I'll have to think of something else I guess.
 
Last thing I read is that AI generated images are not copyrightable: https://arstechnica.com/tech-policy...rtificial-intelligence-cannot-be-copyrighted/
But that's just one judge. It's still something that needs debate, I think.
Yeah, that's why I put in the next part about having a copyright at all - but the bottom line is that anyone else is free to use these images as far as I care. But I'm also in Europe so not completely sure how much the US court rulings concern me, and if EU courts decide differently, will the US courts honor those, and so on and so on.

But basically, none of that matters. If I have copyright, I give anyone here a pass to use them, and if there's no copyright, that's even less of an issue.
 
I have a literal beast of a gaming PC (32 cores/ 64 threads 128gb ram and an RTX 3080 FTW3 🤣… liquid cooled and all the yada yada…
My PC is always evolving … Kinda like my pedal collection.
I’ve never seen this.. Woukd love to generate some? Where do you go to find how to do it?
 
I have a literal beast of a gaming PC (32 cores/ 64 threads 128gb ram and an RTX 3080 FTW3 🤣… liquid cooled and all the yada yada…
My PC is always evolving … Kinda like my pedal collection.
I’ve never seen this.. Woukd love to generate some? Where do you go to find how to do it?
As far as I know, you have two options. I would suggest https://github.com/comfyanonymous/ComfyUI even though it looks a bit scary, the other one is Automatic1111 (https://github.com/AUTOMATIC1111/stable-diffusion-webui). I think the actual ComfyUI installation was quite simple, A1111 was a bit more complicated and I had to do some extra troubleshooting due to only having 8GB of RAM.

After those you need to download the actual models - or at the very least you want the "sd_xl_base_1.0_0.9vae.safetensors" from here https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/tree/main. That's the Stable Diffusion XL 1.0 model. After that I think you should be able to generate something using the default ComfyUI setup, although you might also need "sd_xl_offset_example-lora_1.0.safetensors" from the previous link, and the refiner from here https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/tree/main, I don't remember if those were set up by default or not.

ComfyUI can be scary looking to set up, but the good thing is that the images hold the configurations used to generate that image in them (same goes for A1111, but it's not quite as handy). I used this workflow https://civitai.com/models/111463/s...ing-styles-including-simple-for-an-easy-start for most of those image generations, but you have to install some custom stuff for it to work. The instructions are on that page though, and after that at least the strawberry stuff should open up for you if you load them into ComfyUI and you can see how I did it. Here's some video about it too

I got the rest of the LORA's from Civitai.com too (remember to filter for SDXL so you don't see the outdated stuff). The first ones used this https://civitai.com/models/119526/d...k-w-little-retrowave-graffiti-t-shirt-graphic and the black/red/yellow ones used this https://civitai.com/models/129132/vulcan.

All in all, if you know your way around a computer well and are not afraid of running console commands if needed, plus you want to mess around with generation, and also have the hardware, then it's pretty fun. But if you just want to generate some images, I would suggest just paying for Midjourney or something, since it can be a bit complicated.
 
Copyright means you are ready to foot the bill for a lawyer and go to court should anyone infringe on that copyright. There are no copyright police, you have to do that yourself. I did research on that when I was copyrighting software I wrote. After the research I just said f*** it, I hate courtrooms.... But it looked REALLY COOL to have copyrighted by AK Savage Software, all rights reserved (c) 1988 at the top of a program....
 
Hey, this is a picture of a horse fly that was almost dead and I put it in the sun. What does the AI spit out if you put this in and make it a hell fly or even a strawberry? This dude was about 3/4" long. One BIG nasty fly.....

View attachment 55083

View attachment 55088

View attachment 55089
Tbh I haven't done a lot of image-to-image generation (just text-to-image for the most part), so I don't know exactly the best settings for those. But I did a few generations messing around with it and ended up with these
ComfyUI_temp_qpgnm_00021_.png
Looks like he's sitting on some sort of molten metal that just hardened.
ComfyUI_temp_qpgnm_00023_.png
Fast and bulbous, one might say.
ComfyUI_temp_qpgnm_00022_.png
This fella is hard at work marshalling the other hell flies to their designated landing spots. Built in illuminated wands and everything.
 
Can you try and have the AI generate an image of a haunted moldy moon? or a cheese moon with space rats? dungeon keeper rats? thats all i got. thank you.
 
Little bit confused here. Is it for a VFE Dragon hound clone?
 
Can you try and have the AI generate an image of a haunted moldy moon? or a cheese moon with space rats? dungeon keeper rats? thats all i got. thank you.
I went with a cartoon style since that seemed to be fitting, I can do some more if you want something other.

Haunted moldy moon, here are some of the ones I thought were pretty cool looking.
ComfyUI_temp_qpgnm_00033_.png ComfyUI_temp_qpgnm_00034_.png ComfyUI_temp_qpgnm_00039_.png ComfyUI_temp_qpgnm_00041_.png
As a side note, "haunted" seemed to trigger a LOT of mansions in the generations, I even put "mansion" as a negative prompt to try and decrease them, but they still came through.

Cheese moon with rats, some nice looking stuff, but pretty incoherent, I switched to "space rats walking on a moon made of cheese" halfway so it maybe wouldn't put the moon on the background
ComfyUI_temp_qpgnm_00045_.png ComfyUI_temp_qpgnm_00047_.png ComfyUI_temp_qpgnm_00050_.png
ComfyUI_temp_qpgnm_00059_.png

Special call out to the one with intact text! Only one which had text in it. In case you didn't know, it's pretty rare for AI image generation to get text even halfway correct.
ComfyUI_temp_qpgnm_00046_.png

I'll see about the dungeon keeper rats later on, I have to get some work done too - these were mostly done by clicking the generate button on my second screen while I sit in work teleconferences where they just relay information and I don't need to do anything.

Little bit confused here. Is it for a VFE Dragon hound clone?
The Dragon hound ones are meant for that, yeah. It's the Mad Bean VFE Dragon Hound board. I guess it's a clone since I will build it myself, but I think it's "official" too. Good Vibes is from Moonn Electronics.
 
Tbh I haven't done a lot of image-to-image generation (just text-to-image for the most part), so I don't know exactly the best settings for those. But I did a few generations messing around with it and ended up with these
View attachment 55108
Looks like he's sitting on some sort of molten metal that just hardened.
View attachment 55109
Fast and bulbous, one might say.
View attachment 55110
This fella is hard at work marshalling the other hell flies to their designated landing spots. Built in illuminated wands and everything.
Also a tin teardrop.
 
I went with a cartoon style since that seemed to be fitting, I can do some more if you want something other.

Haunted moldy moon, here are some of the ones I thought were pretty cool looking.
View attachment 55112View attachment 55113View attachment 55114View attachment 55115
As a side note, "haunted" seemed to trigger a LOT of mansions in the generations, I even put "mansion" as a negative prompt to try and decrease them, but they still came through.

Cheese moon with rats, some nice looking stuff, but pretty incoherent, I switched to "space rats walking on a moon made of cheese" halfway so it maybe wouldn't put the moon on the background
View attachment 55116View attachment 55117View attachment 55118
View attachment 55120

Special call out to the one with intact text! Only one which had text in it. In case you didn't know, it's pretty rare for AI image generation to get text even halfway correct.
View attachment 55119

I'll see about the dungeon keeper rats later on, I have to get some work done too - these were mostly done by clicking the generate button on my second screen while I sit in work teleconferences where they just relay information and I don't need to do anything.


The Dragon hound ones are meant for that, yeah. It's the Mad Bean VFE Dragon Hound board. I guess it's a clone since I will build it myself, but I think it's "official" too. Good Vibes is from Moonn Electronics.
Im liking the astronaut mouse!
 
Back
Top