Optimized settings for various stable diffusion models
Added 2024-03-03 23:51:15 +0000 UTCA few days ago I put together a video showcasing the differences in output from 10 different stable diffusion models. My idea was to keep the settings the same between image generations. This was actually flawed because the guidance scale and inference steps need to be optimized for each model in order to get the highest quality out of them.
So, I spent several hours today tuning the guidance scale, inference steps and scheduler for each of the 10 models I tested. I'll put together a video tomorrow, but the optimized default settings are all live right now on https://pixeldojo.ai . Simply enter a prompt, then select a model and the advanced settings will all change to my optimized defaults.
On top of that, I also lowered the Unlimited Plan from $20 to just $5 per month. My hope is to get this technology into the hands of as many people as possible, and hopefully the lower price helps achieve that.
Here are the settings I found success with. Of course I will continue to tweak and tune, but these are a solid starting point:
Stable Diffusion XL
scheduler: K_EULER
Steps: 30
Guidance Scale: 7
Proteus V2
scheduler: K_EULER
Steps: 30
Guidance Scale: 7
SSD 1B
scheduler: K_EULER
Steps: 20
Guidance Scale: 13
Playground V2
scheduler: K_EULER
Steps: 30
Guidance Scale: 2
Juggernaut XL V8
scheduler: K_EULER
Steps: 30
Guidance Scale: 7
Juggernaut XL V9
scheduler: K_EULER_ANCESTRAL
Steps: 30
Guidance Scale: 1
Animagine XL
scheduler: K_EULER
Steps: 50
Guidance Scale: 12
Kandinsky 2.2
scheduler: KarrasDPM
Steps: 50
Guidance Scale: 4
RealvisXL V4
scheduler: K_EULER_ANCESTRAL
Steps: 30
Guidance Scale: 4
DreamshaperXL Turbo
scheduler: K_EULER_ANCESTRAL
Steps: 10
Guidance Scale: 2
Hope this helps, as always, thank you for your support!
