[Bug]: beta 스케줄 사용시 얼굴을 아예 새로 그립니다

[kakaruYH03]

Describe the bug

beta 스케줄 사용시 디테일러가
디노이징을 엄청 높게한거처럼 얼굴을 새로 그리는 현상

Steps to reproduce

사용한 모델은 ntrmix 4.0 + a111 webui 디테일러는 체크박스만 체크한 기본셋팅
다른 스케줄을 사용시에는 이런 현상이 안생기는데 오일러+beta 스케줄+디테일러 사용시 해당 현상이 생깁니다
디노이징을 0.4에서 0.2로 확 낮추면 현상이 덜해지긴 하지만 디노이징 낮춘만큼 디테일이 죽고
다른 스케줄에서는 전부 0.4로해도 잘나왔는데
이상하게 beta 스케줄은 0.4 부터 너무 극단적으로 다르게 나오네요
아니면 beta 스케줄 자체의 문제라 그냥 강도를 낮추고 사용해야하는건지 궁금합니다

Screenshots

오일러+beta 스케줄 사용

dpm++2m +beta 스케줄 사용

Console logs, from start to end.

venv "E:\webSDXL\venv\Scripts\Python.exe"
Python 3.10.11 (tags/v3.10.11:7d4cc5a, Apr  5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Version: v1.10.1
Commit hash: 82a973c04367123ae98bd9abdf80d9eda9b910e2
Installing None
Launching Web UI with arguments: --xformers --xformers-flash-attention --opt-channelslast --enable-insecure-extension-access
2024-12-13 21:50:31.714501: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
2024-12-13 21:50:32.787277: I tensorflow/core/util/port.cc:113] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable `TF_ENABLE_ONEDNN_OPTS=0`.
[-] ADetailer initialized. version: 24.11.1, num models: 12
ControlNet preprocessor location: E:\webSDXL\extensions\sd-webui-controlnet\annotator\downloads
2024-12-13 21:50:39,734 - ControlNet - INFO - ControlNet v1.1.455
Loading weights [3bc9e8fbe6] from E:\webSDXL\models\Stable-diffusion\ntrMIXIllustriousXL_v40.safetensors
Creating model from config: E:\webSDXL\repositories\generative-models\configs\inference\sd_xl_base.yaml
E:\webSDXL\extensions\stable-diffusion-webui-two-shot\scripts\two_shot.py:130: GradioDeprecationWarning: The `style` method is deprecated. Please set these arguments in the constructor instead.
  visual_regions = gr.Gallery(label="Regions").style(grid=(4, 4, 4, 8), height="auto")
E:\webSDXL\extensions\stable-diffusion-webui-two-shot\scripts\two_shot.py:130: GradioDeprecationWarning: The 'grid' parameter will be deprecated. Please use 'columns' in the constructor instead.
  visual_regions = gr.Gallery(label="Regions").style(grid=(4, 4, 4, 8), height="auto")
2024-12-13 21:50:40,641 - ControlNet - INFO - ControlNet UI callback registered.
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
Startup time: 25.3s (prepare environment: 8.9s, import torch: 4.2s, import gradio: 1.4s, setup paths: 5.6s, initialize shared: 0.3s, other imports: 0.6s, load scripts: 2.9s, create ui: 0.8s, gradio launch: 0.3s).

List of installed extensions

adetailer
sd-webui-bmab
sd-webui-controlnet
sd-webui-kohya-hiresfix
sd-webui-lora-block-weight
sd-webui-regional-prompter
sd-webui-supermerger
stable-diffusion-webui-localization-ko_KR
stable-diffusion-webui-two-shot

[dosubot]

The issue you're experiencing seems to be related to the use of the beta schedule with the Euler a sampler in the adetailer extension, which results in the face being redrawn as if the denoising strength is excessively high. This behavior does not occur with other schedules, indicating a specific interaction between the beta schedule and the adetailer extension.

To address this, you might want to experiment with adjusting the "Inpaint denoising strength" parameter in the adetailer settings. This parameter controls the intensity of denoising, and its default value is 0.4, but it can be set between 0.0 and 1.0. Additionally, consider using the "dynamic denoise strength" feature, which adjusts the denoising strength based on the size of the bounding box. This feature is controlled by the ad_dynamic_denoise_power option, which can be set between -10 and 10, with 0 disabling it ^[1][2].

If adjusting these settings does not resolve the issue, it may be worth exploring other samplers or schedules that do not exhibit this behavior.

To continue talking to Dosu, mention @dosu.

---

^{Help Dosu learn! Give it feedback: Great Response | Irrelevant Answer | Incorrect Sources | Too Verbose | Hallucination | Other}

[Bing-su]

말씀하신 게 맞을겁니다. 사용하는 샘플러, 스케줄러 등에 따라 잘 나오는 값을 찾아야 하는 것이 맞습니다.

[bkrew]

I made that experience too, some sampler react differently on the denoise strength, some needed values around 0.2 to start making sense, others start working at 0.5 or 0.6.