Update README.md
Browse files
README.md
CHANGED
|
@@ -81,6 +81,18 @@ For extra information, please read the card link above.
|
|
| 81 |
|
| 82 |
As Illustrious works differently, and the dataset did not have many humanoids, this version works best with animals and robots. YET, you can generate humans if you add "human" in the prompt as seen in the sample.
|
| 83 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 84 |
|
| 85 |
---
|
| 86 |
License: CC BY 4.0
|
|
|
|
| 81 |
|
| 82 |
As Illustrious works differently, and the dataset did not have many humanoids, this version works best with animals and robots. YET, you can generate humans if you add "human" in the prompt as seen in the sample.
|
| 83 |
|
| 84 |
+
---
|
| 85 |
+
|
| 86 |
+
**Tip of usage**
|
| 87 |
+
|
| 88 |
+
* As per usual, it's not recommended running it on Illustrious 0.1 (neither 1.0 to 2.0,) because the 0.1 will render poor images and the others have a tendency of creating unspeakable horrors, at least here on my end.
|
| 89 |
+
* I always prefer running my image LoRAs on any NoobAI EPSILON. Then, if possible use those as base model.
|
| 90 |
+
* If you set `(lowres:1.x)` in the negatives, you can lower the number of steps and maybe lower the weight of this LoRA to 0.8. But it seems to work best at 0.9 ~ 1.0.
|
| 91 |
+
* Illustrious does not work as SDXL, then the "trigger words" can be added in any part of the prompt to possibly make the LoRA work the way it was trained for.
|
| 92 |
+
* Without "trigger words", it still can perform other effects, such as adding that nice plastic/glass bubble, which in case it is much stronger on Illustrious than on SDXL.
|
| 93 |
+
* You can use in the prompt `"a mini [enter landscape here]"` before `"plastic circle"`, and it is possible that the mini "background" will be rendered only inside the "bubble", as you can check in the samples above.
|
| 94 |
+
|
| 95 |
+
|
| 96 |
|
| 97 |
---
|
| 98 |
License: CC BY 4.0
|