How were the Release Note examples generated?
How cherry-picked are the examples in the "Release Notes"? Can you provide me all the generation hyperparameters, including the seed? Thanks.
I used the exact same prompt, that was provided in the example
Positive:
1girl, aqua eyes, baseball cap, blonde hair, closed mouth, earrings, green background, hat, hoop earrings, jewelry, looking at viewer, shirt, short hair, simple background, solo, upper body, yellow shirt,
(waifu, anime, exceptional, best aesthetic, new, newest, best quality, masterpiece, extremely detailed:1.2)
Negative:
lowres, ((bad anatomy)), ((bad hands)), text, missing finger, extra digits, fewer digits, blurry, ((mutated hands and fingers)), (poorly drawn face), ((mutation)), ((deformed face)), (ugly), ((bad proportions)), ((extra limbs)), extra face, (double head), (extra head), ((extra feet)), monster, logo, cropped, worst quality, jpeg, humpbacked, long body, long neck, ((jpeg artifacts)), deleted, old, oldest, ((censored)), ((bad aesthetic)), (mosaic censoring, bar censor, blur censor)
I tried Euler, Euler a and DPM++ 2M with maximum number of steps. I tried adding the wdgoodprompt/wdbadprompt embeddings. I tried different CFG Scales. I tried starting with 384x512 and then Latent Higres. fix to 768x1024 with different Denoising strengths. I tried with and without CLIP skip.
I am using the kl-f8-anime2.ckpt VAE (I even manually force selected it in the settings). I am on the latest version of stable-diffusion-webui at the moment (commit b08500ce, v1.2.0).
What am I doing wrong?
Expected result (from the Release Notes)
Actual results
As you can see, it's not even close.
The release notes you are thinking of are on the beta2 model using the aesthetic version. The generations were too long ago for me to find the exact seed, but they weren't very cherry picked. If you use the aesthetic model you should get a similar result
Sorry, I posted under the wrong repo. The results I provided were generated with wd-1-5-beta2-fp32.yaml, so I was using beta2, just not the aesthetic version.
I'll try using the aesthetic version. I guess, I must have missed the small "images generated with the aesthetic model" in italics under the examples.