The resulting image is heavily darkened

[SaymanDotNet]

Hello.
Downloaded, assembled, launched. Specified the path to the safetensors model. Entered a text task. Got the picture. But the picture always turns out (regardless of the request) darkened. The model does not affect the result.
Compiled with VS2022, Windows 10 22H2 (19045.5537).
The result should definitely be lighter.

[IntptrMax]

Hello. Downloaded, assembled, launched. Specified the path to the safetensors model. Entered a text task. Got the picture. But the picture always turns out (regardless of the request) darkened. The model does not affect the result. Compiled with VS2022, Windows 10 22H2 (19045.5537). The result should definitely be lighter.

Maybe you can use a different vae, or use a different sampler.
Please tell me the model name, vae name and sampler, cfg, steps you have used, I will try it then.

[SaymanDotNet]

model: cyberrealistic_v41BackToBasics.safetensors
from civitai
VAE path is empty.
steps: 30, cfg: 7.0 or 8.0
prompt (from source): realistic, best quality, 4k, 8k, trees, beach, moon, stars, boat,
negative prompt: 2d, 3d, cartoon, paintings
or prompt: car
cat, dog, etc - darkened result
sampler from source.

[SaymanDotNet]

I tried different files with models and different queries, including negative ones. The result is always the same - the picture is very dark. There is a suspicion that somewhere the conversion of 16bit to 32bit is not performed correctly. it's dark from here

[IntptrMax]

Maybe there are some differences between my sampler and the diffusers, It will take some time to fix.

[niksalyan]

I think i found the issue
In SDModel.cs
instead of
image = ((image + 0.5) * 255.0f).clamp(0, 255).@byte().cpu();
The right math would be

image = (((image + 1f) / 2f) * 255.0f).clamp(0, 255).@byte().cpu();

[IntptrMax]

I think i found the issue In SDModel.cs instead of image = ((image + 0.5) * 255.0f).clamp(0, 255).@byte().cpu(); The right math would be

image = (((image + 1f) / 2f) * 255.0f).clamp(0, 255).@byte().cpu();

You are right. Thanks. I have updated the code.
But I think there are still some bugs, it will get some black blocks on the image.