Ok... I spent too much of my work day today playing around with installing SD locally (Running Windows 10 and a GTX1080), so, I'm going to share a consolidated list of tips/guides I found in this thread and elsewhere to help anyone else.
Installation:
I found
videos the best guides (Thanks
bhd2). He takes you through installing Automatic1111 WebUI for Stable Diffusion and gets you up and running fairly quickly. There are some additional dependencies needed, but he takes you through and the links are in the description so anyone with a basic understanding should be able to walk through it:
- Install Automatic1111 WebUI and Stable Diffusion 1.5
Then, Watch these:
- Install SD 2.0
- Install SD 2.1
However, there are some NSFW Filters on 2.1 I found, so thats where the additional models come in.
Models:
The reason I mentioned watching and installing the additional SD Versions, is it goes through again downloading and setting up the .ckpt files and switching to another checkpoint.
These are what are mentioned above; HassanBlend, URPM, SimpMaker-3K1, Dreamlike Photoreal 2.0
And you can find them, as well as more models and resources here:
(Basically GitHub for AI)
Search and download the relevant file and MAKE SURE YOU READ THE DETAILS. Some aren't as simple and just placing the .ckpt file in the directory and require an additional step or two (although Dreamlike did not).
Settings:
This is the part that there is no golden rule for. I played around, followed the below as a bit of a guide and looked at some of the examples available to view on sites like
Prompts:
Again, play around with keywords, look online, through the forum and experiment. Negative prompts are just as important. I have found in my few hours of playing, that I kind of have a set of negative prompts that stay there, then I tune the prompts accordingly. Thanks in big to the below who inspired my prompts:
Brackets:
Dug around, and this is what is meant when you put brackets around something:
(word) - increase attention to word by a factor of 1.1
((word)) - increase attention to word by a factor of 1.21 (= 1.1 * 1.1)
[word] - decrease attention to word by a factor of 1.1
(word:1.5) - increase attention to word by a factor of 1.5
(word:0.25) - decrease attention to word by a factor of 4 (= 1 / 0.25)
\(word\) - use literal () characters in prompt
So... Hopefully putting some of this in a single post helps some of you! I have a running .txt file of notes, prompts and stuff that I've taken from here to help as I play around more with it.
Thanks to everyone else in the thread, Wouldn't have got here without you! Myself, Selena Gomez, Emma Watson and Jessica Alba all thank you!
Jessica Alba, ((face)), (full body), (naked:2), showing head, photorealistic, paparazzi, (Breasts:2) nipples, vagina, smiling, symmetrical face, sitting with legs spread, facing camera, masterterpiece photography, RAW, onlyfans, gorgeous, (highly detailed nipples), 8k, detailed eyes, high-quality breasts, exposed, perky, natural, perfect hands, perfect eyes, waist, legs
Negative prompt: ugly, tiling, poorly drawn hands, poorly drawn, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft, duplicate, coppy, multi, two faces, disfigured, kitsch, oversaturated, grain, low-res, mutation, mutated, extra limb, missing limb, floating limbs, disconnected limbs, malformed hands, blur, out of focus, long neck, long body, disgusting, childish, mutilated, mangled, old, heterochromia, dots, bad quality, weapons, draft
Steps: 25, Sampler: DPM2, CFG scale: 7, Seed: 3408587866, Size: 512x712, Model hash: fc52756a74, Model: dreamlike-photoreal-2.0