[Bug]: Details / saturation / quality loss with hypertiling and default values

Nobody has claimed this yet.

### Checklist

- [ ] The issue exists after disabling all extensions
- [ ] The issue exists on a clean installation of webui
- [X] The issue is caused by an extension, but I believe it is caused by a bug in the webui
- [X] The issue exists in the current version of the webui
- [X] The issue has not been reported before recently
- [ ] The issue has been reported before but has not been fixed yet

### What happened?

**Firstly, I do not know if this is an issue with hypertiling itself or the forge implementation, and I couldn't find that much info on it, so that's why I decided to open an issue here. Or maybe it behaves as intended and I m just an idiot and need to adjust settings properly.**

So, I noticed my images' quality was bad (less details, less saturation) after enabling hypertiling. The default settings are: Tile size (t) = 256, Swap size (s) = 2, Depth = 0

When enabling hypertiling, with the default settings, the image seems to suffer quite the quality loss, issue more apparent with hires.fix enabled.

So here's what I did. I generated an image at 768x512 and another one at 640x640. I then generated another set, with hires.fix enabled and then proceeded to enable Hypertiling and increase the tile size (and for the first gen swap size too).

Here is what I noticed:

- With the default settings, hypertile enabled with t=256, s=2, hires.fix enabled or disabled still presented a noticeable loss in quality. (examples 1 and 2)

- _Without hires fix,_ increasing the tile size resulted in a good image, e.g. from 256 to 384 the issue was not an issue anymore. Going higher did not seem to matter.

- _With hires.fix_, increasing tile size a bit helped, e.g. from 256 to 384 to 512, 768 (quality increases with tile size, up to a point and increasing it so much defeats the point of using hypertiling in the first place).

- Quality / details seem to increase slightly with swap size.

- Depth did not seem to affect anything, although I have not tested it much, I believe I changed it to 2,3,4 at one point, but did not notice anything.

Here's a few examples:


- I messed up the name on 2nd gen (last example) - 2nd item in the dropdown should be NO Hypertiling, WITH HiRES - image on the left side

In the 2nd image for example, you can see how hypertiling enabled, with the default values, produces a worse image than no hypertiling (no hires fix used in either)

Using other resolutions or checkpoints doesn't seem to yield different results.

You can see all examples below.

https://imgsli.com/MjQ2ODcw/1/10
https://imgsli.com/MjQ2ODcy/1/4

### Steps to reproduce the problem

1. Generate an image

2. Generate the same image with hypertiling default values

At this point, a slightly different image is expected, but of the same quality - more or less.

3. Generate the same image with hires.fix, **no** hypertiling this time - I used 4x-Ultrasharp

4. Generate the same image with hires.fix, **with** hypertiling

### What should have happened?

Did not expect such a loss of image quality with the default settings.

### What browsers do you use to access the UI ?

Brave

### Sysinfo

[sysinfo-2024-03-12-19-56.json](https://github.com/lllyasviel/stable-diffusion-webui-forge/files/14577970/sysinfo-2024-03-12-19-56.json)

### Console logs

```Shell
Not sure where the console log is saved, but it looked normal, there's no errors.
```

### Additional information

_No response_

No source file, test, or entry point is named in the issue. Start by reproducing the four image-generation comparisons with default hypertiling values, hires.fix, and the listed resolutions; inspect the hypertiling and hires.fix integration to determine whether the observed quality loss is expected, then add a regression check if a defect is isolated.

Written by the indexing model from the issue text.