File size: 2,820 Bytes
3dabe4a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
# SD Webui ReSharpen
This is an Extension for the [Automatic1111 Webui](https://github.com/AUTOMATIC1111/stable-diffusion-webui), which allows you to increase or decrease the amount of *details* of the generation during the Stable Diffusion pipeline.
> **ie.** This is not just a postprocessing filter
## How to Use
There is simply a slider that controls the "sharpness" of the images, defaulting to 0.
- **Positive** values cause the images to be **noisy**
- **Negative** values cause the images to be **blurry**
> Though, don't actually use values too close 1 or -1 as it will become distorted
For `txt2img`, there is also a second slider for the `Hires. Fix` pass.
You can set positive for the normal pass and negative for the hr pass, or vice versa, for some interesting effects.
There is also now a scaling setting, allowing you to adjust the strength throughout the process *(**eg.** lower the sharpness towards the end)*.
> This Extension is disabled during the [ADetailer](https://github.com/Bing-su/adetailer) phase
**Note:** `Ancestral` samplers *(**eg.** `Euler a`)* do **not** work. Tested and confirmed to work for both `Euler` and `DPM++ 2M Karras`.
## Samples
<table>
<thead align="center">
<tr>
<td>Sharpness</td>
<td><b>-0.5</b></td>
<td><b>-0.25</b></td>
<td><b>0.0</b></td>
<td><b>0.25</b></td>
<td><b>0.5</b></td>
</tr>
</thead>
<tbody align="center">
<tr>
<td>Result</td>
<td><img src="samples\-05.jpg" width=128></td>
<td><img src="samples\-025.jpg" width=128></td>
<td><img src="samples\00.jpg" width=128></td>
<td><img src="samples\025.jpg" width=128></td>
<td><img src="samples\05.jpg" width=128></td>
</tr>
</tbody>
</table>
<p align="center">
<img src="samples/plot.jpg">
</p>
## How does it work?
I initially wrote this Extension to try speeding up the Diffusion process.
Since the latent noises gradually converges to around 0,
I tried to see if I can just add the delta between the current step and the previous step, perhaps multipled by a scaler,
to the current Tensor, in order to help it reaches 0 faster.
However, after fiddling around for a week, I still couldn't get it to consistently produce an actually better result.
So I gave up on the original idea.
Though in the process, I also found out this "sharpening" interaction. If I had to guess,
since the Diffusion is a denoising process, if I add the delta again, then it will overly denoise to create a blurry image;
if I subtract the delta, then it will basically add the noise back to create a noisy image.
## To Do
- [X] Generation InfoText
- [X] Pasting InfoText Support
- [X] X/Y/Z Plot Support
- [X] Scaling Support
|