Update README.md
Browse files
README.md
CHANGED
|
@@ -10,8 +10,22 @@ license: mit
|
|
| 10 |
|
| 11 |
## Description
|
| 12 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
## Model Settings
|
| 14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
## Donations
|
| 16 |
|
| 17 |
- __Tron/USDT (TRC20)__ : `TPXpiWACUZXtUszDbpLeDYR75NQTwngD8o`
|
|
|
|
| 10 |
|
| 11 |
## Description
|
| 12 |
|
| 13 |
+
This is the newest SDXL based model from [Mann-E](https://mann-e.com) platform, which is a generative AI startup based in Iran. This model used thousands of midjourney generated images in order to make it possible to make high-quality images. Also, we've used a lot of tricks in order to make it possible to make the model as fast as SDXL Turbo or any other model which claims to be fast.
|
| 14 |
+
|
| 15 |
+
The model has been mostly developed by Founder and CEO of Mann-E, [Muhammadreza Haghiri](https://haghiri75.com/en) and a team of four. We spent months on collecting the data, labeling them and training this model. The model is _mostly uncensored_ and tested with Automatic1111.
|
| 16 |
+
|
| 17 |
## Model Settings
|
| 18 |
|
| 19 |
+
- CLIP Skip: 1 or 2 are both fine. 1 gives better results.
|
| 20 |
+
- Steps: 6-10. Usually 8 is perfect.
|
| 21 |
+
- CFG Scale: 2-4.
|
| 22 |
+
- Scale: 768x768 and 832x832 are just fine. Higher isn't tested. For 16:9 just try 1080x608
|
| 23 |
+
|
| 24 |
+
## Additional Notes
|
| 25 |
+
|
| 26 |
+
- SDXL 1.0 LoRas are working just fine with the model.
|
| 27 |
+
- ControlNet, IPAdapter, InstantID are just fine.
|
| 28 |
+
|
| 29 |
## Donations
|
| 30 |
|
| 31 |
- __Tron/USDT (TRC20)__ : `TPXpiWACUZXtUszDbpLeDYR75NQTwngD8o`
|