Easy Diffusion 2.5, the simplest one click install I have found
Hello friends
Today I want to share with you the easiest local installation I have found for a Stable Diffusion image generator: this is Easy Diffusion 2.5. You just have to download a simple .exe file (if you have Windows) from their page, but if you have Mac or Linux they are supported too.
the link is :https://github.com/cmdr2/stable-diffusion-ui
In the Windows installation the only thing you have to do (actually there are two things to do) is to run the .exe file, then you have to be careful to install it in a top level directory of your hard drive (or ssd, of course), for example C:\EasyDiffusion (or D:\EasyDiffusion, E:\EasyDiffusion, whatsoever) and the program takes care to install everything Easy Diffusion needs to run. It creates even a shortcut on your desktop if you want (I wanted). The base installation requires 12.5 Gb on your drive, it comes with one model, sd-v1-4 that is a quite outdated model. To add a model just sign to Hugging Face and CivitAi and you will found all the models your heart may desire. You have to download them and put them in the (your drive):\EasyDiffusion\models\stable-diffusion . Be careful because they are quite big and before you use them do a virus scan. In the (your drive):\EasyDiffusion\models\ you will find other directories to put other kinds of models (hypernetworks, VAE, LoRA ) if you want to try to tinker with.
After you have installed properly your copy of Easy Diffusion on your computer it will open a browser page, and this is how it looks:
you may change the look if you don’t like it from settings/theme. If you look carefully there is a negative prompt, which means the things that you do not want in your image. Sometimes the pogram understand them, sometimes don’t, so keep generating images and fine-tune your prompts (both positive and negative).
The image generation settings are here:
You may see various settings, mainly the model does the dirty job and the Vae, hypernetwork and the GFPGANv1.3 (fix incorrect faces and eyes) tweak the image. GFPGANv1.3 and VAE works mainly on the same things. you may choose even not to use them.
There is another panel that makes tweaks on your positive prompt, which is this one:
From here you have a visual clue on how your image can appear:
Now I generate some images:
With an rtx 3060 12Gb tooks 41 seconds to generate 4 images, with a prompt without talent on an out of the box outdated model. With more commitment is possible to generate much much better images. The last words I want to spend in that rendering these images with the cpu takes forever, so if you want to have fun with generated images get a proper graphic card. Even the chat bots run on GPU (well, chat bots are more forgiving). Hope you have fun generating cool stuff
Hugs
Alessandro
