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-V1.1 (Stabilized Edition)
|
| 13 |
+
|
| 14 |
+
**Cosmos V1.1** fixes the tensor explosion issue of V1.0. It moves away from theoretical LLM DARE scaling and uses **Sparse Transfer Learning**.
|
| 15 |
+
|
| 16 |
+
## The Fix
|
| 17 |
+
* **Removed Aggressive Scaling:** We no longer multiply weights by 10x. This eliminates the "deep-fried" noise artifacts.
|
| 18 |
+
* **Sparse Cleaning:** We still **Drop (Prune)** 40-80% of the weak noise data from NoobAI, ensuring only the clean "bone structure" and "core style" are transferred to IllumiYume.
|
| 19 |
+
* **Result:** The sharpness of DARE, without the destruction.
|
| 20 |
+
|
| 21 |
+
## Recommended Settings
|
| 22 |
+
* **Sampler:** Euler a / DPM++ 2M SDE
|
| 23 |
+
* **CFG:** 5.0 - 7.0
|
| 24 |
+
* **Steps:** 28+
|