Stable Diffusion AI (Don't be tricked into paying for what is free) (1 Viewer)

Felldude

Content Creator
Joined
May 3, 2014
DO NOT PAY FOR AI SERVICES

I'll start by saying that AI image technology started around 2017 (It was free and open source then also)
It has advanced rapidly and many websites (malicious or not) have capitalized on this.

Meta, Google, Intel, NVIDIA, Microsoft, AMD all of these company's are working together and releasing their content for free and open source.
One issue with this is it is overwhelming how many programs and resources their are out their.

Probably the most popular uses for AI is image generation.
(Not including AI up scaling which is likely happening right now on your computer)

A collaboration by Facebook (Meta) and other company's lead to the most popular of the programs:


Stable Diffusion (AI Image Generation)

Q. Can anyone run stable diffusion on their computer?

A. Nearly anyone. If you have a computer with an X86 CPU and more then 4GB of ram you can generate 512x512 images
on their PC locally with free open source customizable software. With hundreds of checkpoints to choose from (Realistic to Anime) and
100,000 of thousands of (Lora's) like custom "brushes" for your image.
The ram is the actual gate keeper for the checkpoints. (Most 1.5 are 2GB and under but the new XL SD is over 6GB)
If the math is being handled by the CPU you would need a third generation CPU to run BF16 but that is not a hard requirement.
Any X86 processor can handle FP32 math. (I don't know how far back you would want to try but in theory any X86)



Q. What about those free websites?


A. I'm not commenting on any one site, however the list of free sites doing questionable things out ways the altruistic ones.

I'm sure their are completely safe sites like doing a google colab or the deviant art model that was put out.
And for some people the novelty of it out ways the interest.




Q. What programs are available.

A. Many, but I will focus on two.




The first being Automatic 1111 WebUI



Automatic 1111 WebUI is a good start for beginners and a good choice for NVIDIA or CPU users.
If you plan on using a AMD card I would recommend Comfy UI
For those with old graphics cards the CPU only install may be your best choice.
NVIDIA card users need to make sure and install Pytorch+CUDA, and the CUDNN files



The second being Comfy UI



Comfy UI is the professional setup in many ways but it has a steep learning curve.
NVIDIA card users need to install Pytorch+CUDA
AMD card users will benefit from doing a Direct-ML Onyx conversion to maximize their graphics cards




Currently Civitai host the most checkpoints and lora's,
If you post your content their be aware they claim redistribution and license rights for life


Tools to uncensor games, translate text, identify images and a whole lot of other cool things are nearly all open source.
At most you might have to sign up as a devolper at sites like NVidia.

If you recognize the source of this image shout it out:

Avatar.jpg
 

Killerbeas2

3D Artist
Joined
Nov 21, 2023
I actually recently found Foocus to be better then a1111, mostly because of how easy it is to run. I did comfyui, but found I didn't need the complexity, I was running ideas just to try the tech, no improve my performance.

Recently started training models, and I'm now addicted to it. Main issue is getting good training images. I often will screen caps from all the vids.
 

superfly86

Illustrator
Joined
Sep 1, 2024
hi, can you tell more about AI models available offline ? Maybe some software for decensor/deepfake or video creation?
 

Users who are viewing this thread

Top


Are you 18 or older?

This website requires you to be 18 years of age or older. Please verify your age to view the content, or click Exit to leave.