Upload folder using huggingface_hub
Browse files- README.md +30 -0
- ltx-2-19b-distilled-lora-384.safetensors +3 -0
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# LTX2 Distilled Refine LoRA
|
| 2 |
+
|
| 3 |
+
This repo contains the distilled LoRA used for LTX-2 two-stage refinement.
|
| 4 |
+
FastVideo loads it only when `ltx2_refine_enabled=True` (or the equivalent refine
|
| 5 |
+
flags). The base model repo can reference this LoRA via
|
| 6 |
+
`fastvideo_refine_lora_path` in `model_index.json`, so users do not need to pass
|
| 7 |
+
the LoRA path manually.
|
| 8 |
+
|
| 9 |
+
## Contents
|
| 10 |
+
|
| 11 |
+
- `ltx-2-19b-distilled-lora-384.safetensors` — distilled LoRA weights for the
|
| 12 |
+
stage-2 refine denoising pass.
|
| 13 |
+
|
| 14 |
+
## Usage (FastVideo)
|
| 15 |
+
|
| 16 |
+
Use a distilled LTX-2 diffusers repo that includes
|
| 17 |
+
`fastvideo_refine_lora_path="FastVideo/LTX2-Distilled-LoRA"` in its
|
| 18 |
+
`model_index.json`. Then enable refinement:
|
| 19 |
+
|
| 20 |
+
```python
|
| 21 |
+
from fastvideo import VideoGenerator
|
| 22 |
+
|
| 23 |
+
generator = VideoGenerator.from_pretrained(
|
| 24 |
+
"FastVideo/LTX2-Distilled-Diffusers",
|
| 25 |
+
ltx2_refine_enabled=True,
|
| 26 |
+
)
|
| 27 |
+
```
|
| 28 |
+
|
| 29 |
+
FastVideo will download this LoRA repo automatically only when refinement is
|
| 30 |
+
enabled.
|
ltx-2-19b-distilled-lora-384.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2718f89582003cbb5b616635f18c091641917a3f3e5a2f2ad0fb3d5fdd153534
|
| 3 |
+
size 7674558424
|