updates Junko README to reflect correct training resolution
Browse files- junko/README.md +3 -2
junko/README.md
CHANGED
|
@@ -37,7 +37,7 @@ For her hangry expression: `open mouth, wavy mouth, skin fang, (tearing up, cryi
|
|
| 37 |
- These were reduced because the AI was generating too many "multiple views" images.
|
| 38 |
- 3 batch size, 4 epochs
|
| 39 |
- `(131 * 9 + 9 * 6) / 3 * 4` = 1644 steps
|
| 40 |
-
- 0.
|
| 41 |
- Initially tagged with WD1.4 swin-v2 model. Tags pruned/edited for consistency.
|
| 42 |
- `constant_with_warmup` scheduler
|
| 43 |
- 1.5e-5 text encoder LR
|
|
@@ -45,7 +45,8 @@ For her hangry expression: `open mouth, wavy mouth, skin fang, (tearing up, cryi
|
|
| 45 |
- 1e-5 optimizer LR
|
| 46 |
- Used network_dimension 128 (same as usual) / network alpha 128 (default)
|
| 47 |
- Resized to 32 after training
|
| 48 |
-
- Training resolution
|
|
|
|
| 49 |
- Trained without VAE.
|
| 50 |
- [Dataset can be found on the mega.co.nz repository.](https://mega.nz/folder/SnQDnCRD#ruLvChZGf2vQtWJN87Qs0Q)
|
| 51 |
|
|
|
|
| 37 |
- These were reduced because the AI was generating too many "multiple views" images.
|
| 38 |
- 3 batch size, 4 epochs
|
| 39 |
- `(131 * 9 + 9 * 6) / 3 * 4` = 1644 steps
|
| 40 |
+
- 0.0749 loss
|
| 41 |
- Initially tagged with WD1.4 swin-v2 model. Tags pruned/edited for consistency.
|
| 42 |
- `constant_with_warmup` scheduler
|
| 43 |
- 1.5e-5 text encoder LR
|
|
|
|
| 45 |
- 1e-5 optimizer LR
|
| 46 |
- Used network_dimension 128 (same as usual) / network alpha 128 (default)
|
| 47 |
- Resized to 32 after training
|
| 48 |
+
- Training resolution 768x768.
|
| 49 |
+
- Reduced from 832x832. Junko doesn't really have many very fine details that benefit from the higher resolution, and I think training at 832x832 may negatively impact the quality of images generated at lower resolutions.
|
| 50 |
- Trained without VAE.
|
| 51 |
- [Dataset can be found on the mega.co.nz repository.](https://mega.nz/folder/SnQDnCRD#ruLvChZGf2vQtWJN87Qs0Q)
|
| 52 |
|