Add Readme
Browse files
README.md
ADDED
|
@@ -0,0 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: diffusers
|
| 3 |
+
pipeline_tag: text-to-image
|
| 4 |
+
tags:
|
| 5 |
+
- stable-diffusion-xl
|
| 6 |
+
- v-prediction
|
| 7 |
+
- anime
|
| 8 |
+
- sparse-merge
|
| 9 |
+
license: openrail++
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# Project-Cosmos-V1.4 (Anatomy Anchor)
|
| 13 |
+
|
| 14 |
+
**Cosmos V1.4** is laser-focused on fixing anatomy, specifically fingers and limb proportions. It introduces the "Anchor Method" to eliminate ghosting artifacts.
|
| 15 |
+
|
| 16 |
+
## The Improvements
|
| 17 |
+
* **Middle Block Anchor:** Strength reduced to **0.25**. We are now forcing the critical anatomy layers (fingers/core) to stick 75% to IllumiYume's proven structure. This eliminates the "double finger" or "ghosting" effect caused by conflicting model data.
|
| 18 |
+
* **Hyper-Sparse Input:** Input Block Drop Rate increased to **85%**. By discarding even more "weak" pose data from NoobAI, we ensure that only the most confident and compatible structural updates are applied.
|
| 19 |
+
* **Style Preservation:** Output blocks retain the V1.3 High-Fidelity settings, ensuring the image still looks crisp and detailed.
|
| 20 |
+
|
| 21 |
+
## Recommended Settings
|
| 22 |
+
* **Sampler:** Euler a / DPM++ 2M SDE
|
| 23 |
+
* **CFG:** 5.0 - 7.0
|
| 24 |
+
* **Steps:** 28+
|