Update README.md
Browse files
README.md
CHANGED
|
@@ -3,15 +3,20 @@ license: other
|
|
| 3 |
license_name: faipl-1.0-sd
|
| 4 |
license_link: https://freedevproject.org/faipl-1.0-sd/
|
| 5 |
pipeline_tag: text-to-image
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
---
|
| 7 |
|
| 8 |
-
## Constructing...
|
| 9 |
-
Checkpoint is not available yet. please see this repo later.
|
| 10 |
-
|
| 11 |
# Millennium-IL
|
| 12 |
-
Modified Illustrious-XL-0.1 with Blue Archive style
|
| 13 |
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
## Prompt Guidelines
|
| 17 |
Almost same as the base model
|
|
@@ -20,10 +25,10 @@ Almost same as the base model
|
|
| 20 |
None(Works good without `masterpiece, best quality`)
|
| 21 |
|
| 22 |
## Recommended Negative Prompt
|
| 23 |
-
`worst quality, bad quality, lowres, jpeg artifacts, unfinished`
|
| 24 |
|
| 25 |
## Recommended Settings
|
| 26 |
-
Steps: 14-
|
| 27 |
|
| 28 |
Sampler: DPM++ 2M(dpmpp_2m)
|
| 29 |
|
|
@@ -33,7 +38,7 @@ Guidance Scale: 4-9
|
|
| 33 |
|
| 34 |
|
| 35 |
## Training information
|
| 36 |
-
Finetuned Illustrious-XL-
|
| 37 |
|
| 38 |
- Network module: lycoris_kohya(algo=lora, dora_wd=True)
|
| 39 |
- Resolution: 1024(Bucketing enabled, min 512, max 2048)
|
|
@@ -44,7 +49,9 @@ Finetuned Illustrious-XL-0.1 by repeating the training and merging a DoRA 7 time
|
|
| 44 |
- Noise Offset: 0.04
|
| 45 |
- Immiscible Noise: 2048
|
| 46 |
- Batch size: 1
|
| 47 |
-
|
|
|
|
|
|
|
| 48 |
|
| 49 |
## Dataset information
|
| 50 |
Dataset size: 338
|
|
|
|
| 3 |
license_name: faipl-1.0-sd
|
| 4 |
license_link: https://freedevproject.org/faipl-1.0-sd/
|
| 5 |
pipeline_tag: text-to-image
|
| 6 |
+
base_model:
|
| 7 |
+
- OnomaAIResearch/Illustrious-xl-early-release-v0
|
| 8 |
+
tags:
|
| 9 |
+
- stable-diffusion
|
| 10 |
+
- stable-diffusion-xl
|
| 11 |
---
|
| 12 |
|
|
|
|
|
|
|
|
|
|
| 13 |
# Millennium-IL
|
|
|
|
| 14 |
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
Modified Illustrious-XL-v0.1 with Blue Archive style
|
| 18 |
+
|
| 19 |
+
Do not expect quality!
|
| 20 |
|
| 21 |
## Prompt Guidelines
|
| 22 |
Almost same as the base model
|
|
|
|
| 25 |
None(Works good without `masterpiece, best quality`)
|
| 26 |
|
| 27 |
## Recommended Negative Prompt
|
| 28 |
+
`worst quality, low quality, bad quality, lowres, jpeg artifacts, unfinished, monochrome`
|
| 29 |
|
| 30 |
## Recommended Settings
|
| 31 |
+
Steps: 14-28
|
| 32 |
|
| 33 |
Sampler: DPM++ 2M(dpmpp_2m)
|
| 34 |
|
|
|
|
| 38 |
|
| 39 |
|
| 40 |
## Training information
|
| 41 |
+
Finetuned Illustrious-XL-v0.1 by repeating the training and merging a DoRA 8 times with sd-scripts.
|
| 42 |
|
| 43 |
- Network module: lycoris_kohya(algo=lora, dora_wd=True)
|
| 44 |
- Resolution: 1024(Bucketing enabled, min 512, max 2048)
|
|
|
|
| 49 |
- Noise Offset: 0.04
|
| 50 |
- Immiscible Noise: 2048
|
| 51 |
- Batch size: 1
|
| 52 |
+
- Gradient Accumulation steps: 1 or 2
|
| 53 |
+
- Dim/Alpha: 16/4
|
| 54 |
+
- Conv Dim/Alpha: 1/0.1
|
| 55 |
|
| 56 |
## Dataset information
|
| 57 |
Dataset size: 338
|