SD 1.5 models are no longer working correctly for me. They all generate images that are just noise and usually mostly grey or white.

[ReppiksProductions]

All of my SD 1.5 models are now creating gray and/or static images.

The type of results I'm getting:

I started using SDXL models about a month ago, and all was going great, until I wanted to try to test some prompts and things in my SD 1.5 models. I don't recall really changing anything besides switching to a SDXL model, adding the yaml file, and things started working for SDXL.

Is there a setting I ticked or parameter I added somewhere I'm forgetting about to use 1.5 models again? Is there an argument I need to set to get things working?

Some thing I know is not the problem:
I know it's not the VAE, I've manually selected them and have switched it to auto with no good results.
I've deleted the VENV and repositories folders, and have updated A1111
I've disabled all plugins and still get the weird generations
I finally got xformers working...but doesn't help

Here is the image metadata:
1girl, A candid portrait of a lovely blonde college student, (slim face1:5), (slim cheeks:1.5), skinny, grey eyes

Negative prompt: bad art, ugly, deformed, watermark, duplicated

Steps: 20, Sampler: DPM++ SDE Karras, CFG scale: 7, Seed: 665468701, Size: 512x768, Model hash: 338b85bc4f, Model: juggernaut_reborn, VAE hash: 735e4c3a44, VAE: vae-ft-mse-840000-ema-pruned.safetensors, Version: v1.7.0

What the console looks like:

Loading weights [07919b495d] from E:_python_projects\stable-diffusion-webui\models\Stable-diffusion\picxReal_10.safetensors

2024-02-12 19:22:36,481 - ControlNet - INFO - ControlNet UI callback registered.

Creating model from config: E:_python_projects\stable-diffusion-webui\configs\v1-inference.yaml

Running on local URL: http://127.0.0.1:7860/

To create a public link, set share=True in launch().

Startup time: 51.4s (initial startup: 0.1s, prepare environment: 22.5s, import torch: 6.4s, import gradio: 3.6s, setup paths: 3.2s, initialize shared: 0.5s, other imports: 2.9s, setup codeformer: 0.4s, list SD models: 0.4s, load scripts: 6.1s, initialize extra networks: 0.2s, create ui: 4.6s, gradio launch: 0.8s).

Loading VAE weights specified in settings: E:_python_projects\stable-diffusion-webui\models\VAE\vae-ft-mse-840000-ema-pruned.safetensors

Applying attention optimization: Doggettx... done.

Model loaded in 23.7s (load weights from disk: 3.3s, create model: 0.7s, apply weights to model: 14.2s, load VAE: 4.0s, load textual inversion embeddings: 0.5s, calculate empty prompt: 0.7s).

100%|██████████████████████████████████████████████████████████████████████████████████| 20/20 [00:10<00:00, 1.95it/s]

Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:09<00:00, 2.15it/s]

Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:09<00:00, 2.21it/s]

[ReppiksProductions]

I feel very silly now lol. It was just a syntax error causing this. The "(slim face1:5)" wasn't properly inserted and didn't work on the 1.5 models. It should have been (slim face:1.5) and I probably don't need it that high. I didn't try any prompts without copying this because I was trying to explore creating a consistent character. I still find it a bit strange that SDXL doesn't get affected by this but the SD 1.5 models do.

[remixer-dec]

I had a similar problem (noise output for most samplers), downgrading pytorch to 2.1.2 solved it for me.