Release FlashDelight and FlashRelight inference checkpoints
Browse files- README.md +40 -0
- flashdelight_rank0_full.pt +3 -0
- flashrelight_rank0_full.pt +3 -0
README.md
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model: Wan-AI/Wan2.1-T2V-1.3B
|
| 3 |
+
pipeline_tag: video-to-video
|
| 4 |
+
tags:
|
| 5 |
+
- video-relighting
|
| 6 |
+
- portrait-relighting
|
| 7 |
+
- pytorch
|
| 8 |
+
- arxiv:2611.06355
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# FlashRelight
|
| 12 |
+
|
| 13 |
+
Official inference checkpoints for **FlashRelight: Portrait Video Relighting with Dynamic Lighting** (NeurIPS 2026). [Paper](https://arxiv.org/abs/2611.06355) 路 [Code](https://github.com/vivoCameraResearch/FlashRelight)
|
| 14 |
+
|
| 15 |
+
FlashDelight suppresses the original lighting in a portrait video. FlashRelight uses the delighted clip to synthesize lighting from an HDR environment map or a reference video. Text control works best with lighting descriptions similar to the training captions. The released models process 49 frames at 480 脳 832 pixels.
|
| 16 |
+
|
| 17 |
+
| Checkpoint | Size | Use |
|
| 18 |
+
| --- | ---: | --- |
|
| 19 |
+
| `flashdelight_rank0_full.pt` | 6.93 GB | First stage. Use it on its own for delighting or as the first step in relighting. |
|
| 20 |
+
| `flashrelight_rank0_full.pt` | 7.04 GB | Second stage. Use it with FlashDelight for relighting. |
|
| 21 |
+
|
| 22 |
+
## Download
|
| 23 |
+
|
| 24 |
+
From the FlashRelight code repository, download both checkpoints to the `weights` directory.
|
| 25 |
+
|
| 26 |
+
```bash
|
| 27 |
+
hf download JiahuiSheng/FlashRelight flashdelight_rank0_full.pt flashrelight_rank0_full.pt --local-dir weights
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
For standalone delighting, download only `flashdelight_rank0_full.pt`. Both modes also use the shared Wan2.1 configuration, VAE, text encoder, and tokenizer. [Wan2.1 base files](https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B)
|
| 31 |
+
|
| 32 |
+
<details>
|
| 33 |
+
<summary>SHA-256 checksums</summary>
|
| 34 |
+
|
| 35 |
+
```text
|
| 36 |
+
04925b6ca8a4b537685a2791be65d0a51d2025fba8d2cb4ae17d4bf9530c1f5c flashdelight_rank0_full.pt
|
| 37 |
+
8b9989cdebd2d9815f3678baf4493fbf7c0ae65fe2c74756769e4f4df1c226d7 flashrelight_rank0_full.pt
|
| 38 |
+
```
|
| 39 |
+
|
| 40 |
+
</details>
|
flashdelight_rank0_full.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:04925b6ca8a4b537685a2791be65d0a51d2025fba8d2cb4ae17d4bf9530c1f5c
|
| 3 |
+
size 6932318272
|
flashrelight_rank0_full.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8b9989cdebd2d9815f3678baf4493fbf7c0ae65fe2c74756769e4f4df1c226d7
|
| 3 |
+
size 7044091124
|