3-step Guide Stable diffusion lazy fluffy ass mode

Intro:

Hello this little guide for stable diffusion. If you GPU or IGPU "APU" (released after 2016) it supports vulkan api (you can check it using GPU-Z)and have at least 2048 MB VRAM (for 512x512). No matter if you have NVIDIA or AMD or Intel. It must work on any GPU that support vulkan.

How fast is it?

It depend on GPU clock speed and memory clock speed and memory bandwidth, on cheap GPU (IGPU) based on standard DDR3 or DDR4 memory it slow memory , GDDR memory faster but price higher than standard DDR.

test setup:

i have old pc with specs:
AMD APU A8 FM2 socket iGPU radeon R7 with 2048 MB shared from DDR3 1600 ram,with 8 GB ram, and no external GPU,
res:512x512 10 steps
18-30 sec per iteration, around 3 min 15 second for 1 image
it works faster when no 3D graphics or video playback is used
you can turn your old pc to pr0n gen

i check on my main machine with RTX 3060
1-3 sec per step faster than cuda version!

Let's GO!

Step 1 download koboldcpp

https://github.com/LostRuins/koboldcpp/releases/
download koboldcpp_nocuda.exe
(for NVIDIA users: we use Vulkan is not depend on CUDA)

=======================================================================================================================

Step 2 Download SD weights you can select one of models below or all of them!

(note select FP16 version around 2 GB!)

Indigo Furry mix
size: 2.13 GB
https://huggingface.co/uhralk/Indigo_Furry_mix/blob/main/indigo_Furrymix_v120_hybrid_fin_fp16.safetensors

=======================================================================================================================
Yiffymix v36
this old version but work
size: 2.13 GB
https://huggingface.co/Mistermango24/Yiffymix_V3.6/blob/main/yiffymix_v36.safetensors

also download VAE furception
size: 335MB
https://huggingface.co/RedRocket/furception_vae/blob/main/furception_vae_1-0.safetensors

=======================================================================================================================
Photon realistic model (not furry)
size 2.13 GB
https://huggingface.co/digiplay/Photon_v1/blob/main/photon_v1.safetensors

=======================================================================================================================

Step 3 RUN!

After you download kobold and model
run koboldcpp_nocuda.exe
on opened window in presets: select vulkan (it should show your GPU! If not, check you drivers or change GPUID)
don't touch anything other

after on left list go to tab image Gen
Alt Tag

=======================================================================================================================
on image gen tab

Push browse button and select safetensor model
(if you use yiff mix you also select vae)
push green launch button in bottom right corner
Alt Tag

=======================================================================================================================

After open in browser url http://localhost:5001/sdui/ and you in SDUI!

warning default steps is 30 is too much, set to 10-20 steps
Alt Tag

Happy creations!

THE END!

=======================================================================================================================