commit
Browse files
.ipynb_checkpoints/README-checkpoint.md
DELETED
|
@@ -1,81 +0,0 @@
|
|
| 1 |
-
---
|
| 2 |
-
tags:
|
| 3 |
-
- stable-diffusion
|
| 4 |
-
- stable-diffusion-xl
|
| 5 |
-
---
|
| 6 |
-
|
| 7 |
-
# Nekoray-XL-V0.7
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
## Model Card
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
NekoRay v0.7 is a SDXL checkpoint finetuned on SDXL 1.0, using 1.5m quality-tagged images from a selection of image sites. It aims to be the next Waifu-diffusion, creating generation freedom for the SD community.
|
| 14 |
-
|
| 15 |
-
The project is a WIP, and further checkpoints with enhanced augmentations & more images is currently being developed.
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
Two models trained on the same dataset/hardware/hyperparameters are currently available:
|
| 19 |
-
|
| 20 |
-
- **[fp16mixed](https://huggingface.co/trojblue/nekoray-xl-fulldan-bench-1.5m/blob/main/nekoray-xl-1.5m-fp16mixed_e02.safetensors)**: 2epoch, half-precision
|
| 21 |
-
|
| 22 |
-
- **[32full](https://huggingface.co/trojblue/nekoray-xl-fulldan-bench-1.5m/blob/main/nekoray-xl-1.5m-pdg32_e02.safetensors)**: 1.7epoch (still training), full-precision
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
## Usage:
|
| 26 |
-
|
| 27 |
-
It's recommended to use the *exact* same resolution as specified below, since the original SDXL doesn't perform well out of these resolutions. For prompt danbooru-styled caption is preferred.
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
we use the same aspect ratio as the original SDXL:
|
| 32 |
-
|
| 33 |
-
| Height | Width | Aspect Ratio |
|
| 34 |
-
| ------ | ----- | ------------ |
|
| 35 |
-
| 512 | 2048 | 0.25 |
|
| 36 |
-
| 512 | 1984 | 0.26 |
|
| 37 |
-
| 512 | 1920 | 0.27 |
|
| 38 |
-
| 512 | 1856 | 0.28 |
|
| 39 |
-
| 576 | 1792 | 0.32 |
|
| 40 |
-
| 576 | 1728 | 0.33 |
|
| 41 |
-
| 576 | 1664 | 0.35 |
|
| 42 |
-
| 640 | 1600 | 0.4 |
|
| 43 |
-
| 640 | 1536 | 0.42 |
|
| 44 |
-
| 704 | 1472 | 0.48 |
|
| 45 |
-
| 704 | 1408 | 0.5 |
|
| 46 |
-
| 704 | 1344 | 0.52 |
|
| 47 |
-
| 768 | 1344 | 0.57 |
|
| 48 |
-
| 768 | 1280 | 0.6 |
|
| 49 |
-
| 832 | 1216 | 0.68 |
|
| 50 |
-
| 832 | 1152 | 0.72 |
|
| 51 |
-
| 896 | 1152 | 0.78 |
|
| 52 |
-
| 896 | 1088 | 0.82 |
|
| 53 |
-
| 960 | 1088 | 0.88 |
|
| 54 |
-
| 960 | 1024 | 0.94 |
|
| 55 |
-
| 1024 | 1024 | 1.0 |
|
| 56 |
-
| 1024 | 960 | ... |
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
for prompts the following keywords are appended for better separation of genres:
|
| 60 |
-
|
| 61 |
-
```
|
| 62 |
-
'sensitive-rated', 'questionable-rated', 'explicit-rated'
|
| 63 |
-
```
|
| 64 |
-
|
| 65 |
-
The models are intended to be used as a pretrained checkpoint, and further finetuning is **strongly recommended** for downstream use. For more info 'Finetuning' part below:
|
| 66 |
-
|
| 67 |
-
## Finetuning
|
| 68 |
-
|
| 69 |
-
finetuning on SDXL inherently provides better clarity and reduced 'noisinesss' on higher resolutions compared to sd1.4. We've got some good results for further finetuning on various anime-related subject matters, including but not limited to:
|
| 70 |
-
|
| 71 |
-
- style finetunes (tested on nijijourney images & pvc/figure datasets)
|
| 72 |
-
|
| 73 |
-
- character finetunes (tested on 8 Blue Archive characters)
|
| 74 |
-
|
| 75 |
-
- concept finetunes (tested on nsfw gestures)
|
| 76 |
-
|
| 77 |
-
Actual samples of downstream finetunes will be posted once we get the finetuners' conscents.
|
| 78 |
-
|
| 79 |
-
## License
|
| 80 |
-
|
| 81 |
-
The model (which is still very WIP) is intended to be used as a foundation for various downstream finetunes. The license is under discussion but it would be generally follow Openrail-M agreements.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|