santa hat
deerdeer nosedeer glow
Sign In

guy90's guide to generating with Automatic1111 SD1.5/ SDXL/ Pony

guy90's guide to generating with Automatic1111 SD1.5/ SDXL/ Pony

In this article I will be giving you the tools I use to generate images. Below will include VAEs, Embeddings/ TIs, Extensions, and some other methods I use to generate images. Feel free to use any of these and even recommend anything I may be missing in the comments. (Updated examples coming soon)

Stable Diffusion

To start if you haven't already, please have Automatic1111. If you need help downloading check out this guide from Wrench1815--> https://civitai.com/articles/1722/the-no-way-to-fuck-it-up-this-time-guide-to-installing-auto1111-or-forge-or-the-proompt-ninja

If you want to try an experimental version, get Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge

Forge is a fork of automatic1111 that can speed up generation times, especially for those on lower end pcs. It also has features that Automatic1111 does not have built in unless you download extensions. There will also an extension I will mention later that only works on Forge. I personally like using forge, and even with a 4070, my generations are still faster than automatic1111.

VAE

A VAE will help dictate the colors, details and also the compression used in generated images. If you are not using a VAE or a checkpoint without a VAE, the colors will look very bland/ less vibrant. See Examples:

SDXL

I do not use SDXL 1.0 so i can't really speak about what vae to use, however I use Pony. Pony doesn't need a vae in my opinion, however if you have a lower end computer, try using this one to help speed up generation times. Images may look a little less detailed but It will help with generation times. VAE: https://civitai.com/models/140686

If you are using Forge, do not use this as it may change colors that are not intended (Don't know why, forge just doesn't like it).

SD 1.5

I recommend using this VAE: https://civitai.com/models/276082

I don't touch SD 1.5 anymore so there may be others that are better for when I did use it, I used this.

Here is an image without a VAE. See how boring it looks with it's washed out colors.

Here is the same generation but with the VAE. The image is much more vibrant and exciting to be looking at.

Upscaler

Upscalers help make the images a higher resolution when using Hires. fix, however they work good when upscaling images using the extras tab.

Here are two options for 2D generations.

4x foolhardy Remacri: https://huggingface.co/FacehugmanIII/4x_foolhardy_Remacri/blob/main/4x_foolhardy_Remacri.pth

4x-AnimeSharp: https://openmodeldb.info/models/4x-AnimeSharp

4x foolhardy Remacri I personally like using for 2D animations however 4x-AnimeSharp is a good alternative but it primarily for anime generations.

I personally like using these 2 for 3D generations

RealSR DF2K: https://openmodeldb.info/models/4x-realsr-df2k

4xLSDIR: https://openmodeldb.info/models/4x-LSDIR

RealSR DF2K I like to use this one as if has a more realistic/ detailed look, but if you like a more toony look you can use 4xLSDIR without really loosing any quality.

Here is a website for more options: https://openmodeldb.info/

Embeddings (SD 1.5 Only)

I only use this for SD 1.5. I personally don't like using these for SDXL/ Pony.

I recommend Amazing Embeddings - fcNegative + fcPortrait suite: https://civitai.com/models/81575?modelVersionId=97691

There are other versions as well for specific things like portraits, however the one linked works for all stuff. This helps add more detail to your generations.

Here is an image without fcNegative. You can see there is a low amount of detail, especially in the background with certain things not connecting.

Here is the same seed but with fcNegative. You can tell that there is more detail in the image and no weird generations

Extensions

Adetailer

https://github.com/Bing-su/adetailer

This extension will auto in paint onto your generated image. This can fix things like faces, hands, eyes, even bodies if you have multiple. There are also other Adetailer models you can find that are trained specifically on other things. Try them out and see how you like them.

This is an image with no adetailer at a resolution of 512x768. I did this to help see the difference better.

Here is an image with adetailer with the same resolution. You will see the face looks better

Adetailer Models I use

I personally like using this one for faces: https://civitai.com/models/207406/adetailer-face-finder-furry-anime-realistic-ddetailer

Regional Prompter

https://github.com/hako-mikan/sd-webui-regional-prompter

Want to generate multiple characters in an image, try Regional Prompter. I'm not going to go over it as i don't really use it but check out this guide from FallenIncursio on how to use it: https://civitai.com/models/339604?modelVersionId=380265

Forge Attention Couple

https://github.com/Haoming02/sd-forge-couple

I personally like using this one instead of Regional Prompter however this can only be used in Forge. This one is basically much simpler and in my opinion I think it generates much better results.

CD Tuner

https://github.com/hako-mikan/sd-webui-cd-tuner

This one I have recently started using so I personally am not that good with it, but I really like the results I get with it. For a more in depth look, check out PotatCat's article: https://civitai.com/articles/5776/potats-collection-of-useful-extensions-and-other-stuff#cd(colordetail)-tuner-gof9j09v3

Upscaling methods

There are 2 ways that I upscale

Hi-rez fix and this method: https://civitai.com/articles/4560/upscaling-images-using-multidiffusion

Conclusion

Again these are what I am currently using to generate and if anyone has any recommendations, feel free to drop them down in the comments. Also if there are any questions feel free to message me or in the comments.

361

Comments