After having updated Stable Diffusion after a long time of no use, I can no longer recreate the images I once made.

[McSnackAI]

As the title says. The UI and more options have been added since the last time I could create the good images.

This is the images I once created:

This is the recreation after a couple of updates later with the "same" settings:

parameters

(a girl in steampunk fantasy world), (ultra detailed prosthetic arm and leg), (light particles ,colorful magic:1.2), blueprints, (potions), mechanical tools, plants, (a small cat pet:1.1), full body,
BREAK
(an overview of medieval fantasy city 1.2), (steampunk fantasy art:1.2), outdoors, (Steam-powered machines:1.1), (clockwork automatons:1.1), (a small wooden toy), (intricate details:1.5), colorful details, iridescent colors
BREAK
((masterpiece, best quality)), absurdres, highres, 4k, ultra detailed, solo, (asymmetry:1.1), sitting, detailed lighting, inspired by Hayao Miyazaki, art by Katsuhiro Otomo, art by Hiromu Arakawa, art by Hiroshi Yoshida
Negative prompt: (EasyNegative:1.2), (worst quality, low quality:1.4), (blurry:1.4)

Steps: 40, Sampler: DPM++ 2M Karras, CFG scale: 30, Seed: 3692999247, Size: 896x512, Model hash: 86bd0c547c, Model: cardosAnime_v20, Denoising strength: 0.6, Clip skip: 2, ENSD: 31337, Version: v1.2.1, Hires upscale: 2, Hires steps: 40, Hires upscaler: Latent (bicubic antialiased), Dynamic thresholding enabled: True, Mimic scale: 9, Threshold percentile: 100, Mimic mode: Half Cosine Up, Mimic scale minimum: 3, CFG mode: Half Cosine Up, CFG scale minimum: 3

I have version: Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)]
Version: v1.10.1

[McSnackAI]

Im heart broken...

Let me know if you want more details that I can provide

[missionfloyd]

The model hashes are different. You might try using the unpruned version, as you were before.

The VAE info is missing from the "before" image. Are you using the same VAE?

You could also try the options in Settings > Compatibility, but they should be applied automatically.
https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Seed-breaking-changes

Your dynamic thresholding settings also appear to be different.

If all else fails, you can downgrade with git checkout v1.2.1 or git reset --hard v1.2.1 (and delete the venv folder).

[McSnackAI]

I also get as override settings: Downcast alphas_cumprod: true

[McSnackAI]

I will check your recommendations after work