Instructions to use Lightricks/LTX-2-Marking with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- AudioSeal
How to use Lightricks/LTX-2-Marking with AudioSeal:
# Watermark Generator from audioseal import AudioSeal model = AudioSeal.load_generator("Lightricks/LTX-2-Marking") # pass a tensor (tensor_wav) of shape (batch, channels, samples) and a sample rate wav, sr = tensor_wav, 16000 watermark = model.get_watermark(wav, sr) watermarked_audio = wav + watermark# Watermark Detector from audioseal import AudioSeal detector = AudioSeal.load_detector("Lightricks/LTX-2-Marking") result, message = detector.detect_watermark(watermarked_audio, sr) - Notebooks
- Google Colab
- Kaggle
Add LTX-2 marking pack (PixelSeal + AudioSeal)
Browse files- LICENSE-audioseal +21 -0
- LICENSE-videoseal +21 -0
- README.md +128 -0
- audioseal_detector.pth +3 -0
- audioseal_generator.pth +3 -0
- pixelseal.pth +3 -0
- pixelseal.yaml +40 -0
- videoseal_attenuation.yaml +4 -0
LICENSE-audioseal
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
MIT License
|
| 2 |
+
|
| 3 |
+
Copyright (c) Meta Platforms, Inc. and affiliates.
|
| 4 |
+
|
| 5 |
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 6 |
+
of this software and associated documentation files (the "Software"), to deal
|
| 7 |
+
in the Software without restriction, including without limitation the rights
|
| 8 |
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 9 |
+
copies of the Software, and to permit persons to whom the Software is
|
| 10 |
+
furnished to do so, subject to the following conditions:
|
| 11 |
+
|
| 12 |
+
The above copyright notice and this permission notice shall be included in all
|
| 13 |
+
copies or substantial portions of the Software.
|
| 14 |
+
|
| 15 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 16 |
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 17 |
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 18 |
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 19 |
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 20 |
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
| 21 |
+
SOFTWARE.
|
LICENSE-videoseal
ADDED
|
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
MIT License
|
| 2 |
+
|
| 3 |
+
Copyright (c) Meta Platforms, Inc. and affiliates.
|
| 4 |
+
|
| 5 |
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
| 6 |
+
of this software and associated documentation files (the "Software"), to deal
|
| 7 |
+
in the Software without restriction, including without limitation the rights
|
| 8 |
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
| 9 |
+
copies of the Software, and to permit persons to whom the Software is
|
| 10 |
+
furnished to do so, subject to the following conditions:
|
| 11 |
+
|
| 12 |
+
The above copyright notice and this permission notice shall be included in all
|
| 13 |
+
copies or substantial portions of the Software.
|
| 14 |
+
|
| 15 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
| 16 |
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
| 17 |
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
| 18 |
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
| 19 |
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
| 20 |
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
| 21 |
+
SOFTWARE.
|
README.md
ADDED
|
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
library_name: pytorch
|
| 4 |
+
tags:
|
| 5 |
+
- watermarking
|
| 6 |
+
- video
|
| 7 |
+
- audio
|
| 8 |
+
- videoseal
|
| 9 |
+
- pixelseal
|
| 10 |
+
- audioseal
|
| 11 |
+
- ltx-2
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# LTX-2 Marking Pack (VideoSeal / AudioSeal)
|
| 15 |
+
|
| 16 |
+
Weights and configs for the opt-in, post-hoc invisible marking used by [LTX-2](https://github.com/Lightricks/LTX-2): **PixelSeal** (VideoSeal family) for video frames and **AudioSeal** for stereo PCM.
|
| 17 |
+
|
| 18 |
+
This repository contains **weights only**. The code that loads, embeds, and detects lives in the public LTX-2 repository under [`packages/ltx-pipelines/src/ltx_pipelines/marking`](https://github.com/Lightricks/LTX-2/tree/main/packages/ltx-pipelines/src/ltx_pipelines/marking). Full documentation is in [`packages/ltx-pipelines/docs/marking.md`](https://github.com/Lightricks/LTX-2/blob/main/packages/ltx-pipelines/docs/marking.md).
|
| 19 |
+
|
| 20 |
+
> The marks are **not** cryptographic signatures. They are a robust, invisible signal that a clip was produced by an LTX-2 pipeline with marking enabled. Absence of a detected mark does not prove a clip was not generated.
|
| 21 |
+
|
| 22 |
+
## Files
|
| 23 |
+
|
| 24 |
+
The LTX-2 loader (`create_markers`) requires all five model files to be present in one directory.
|
| 25 |
+
|
| 26 |
+
| File | Description | SHA-256 |
|
| 27 |
+
|:--|:--|:--|
|
| 28 |
+
| `pixelseal.pth` | PixelSeal embedder + extractor weights (model state only) | `97cff15f75515259b49a2cffc8fcd600c287cafe3f29c1d5c6ef9c81b492c088` |
|
| 29 |
+
| `pixelseal.yaml` | PixelSeal model card (architecture, 256-bit payload, 256 px processing size) | `96c4a448f7d81810780e5bde5a51b039db544b7ec2baf65a605302c1b17362c9` |
|
| 30 |
+
| `videoseal_attenuation.yaml` | JND attenuation channel map referenced by the card | `e0d71354f8945cf600bd4ebd5e2641a9c879e017548dd6cc8a6de10239d7e821` |
|
| 31 |
+
| `audioseal_generator.pth` | AudioSeal 16-bit generator (embed) | `7a845b5fbe9364a63a3909d8ab3fe064d13a76ae4c2e983573e08c69b7b51748` |
|
| 32 |
+
| `audioseal_detector.pth` | AudioSeal 16-bit detector | `8a78e8a83584113523e161fc599fcab10fd0e94c04d2eb9d2fa1e9ec91ab69d9` |
|
| 33 |
+
| `LICENSE-videoseal`, `LICENSE-audioseal` | Upstream MIT licenses | |
|
| 34 |
+
|
| 35 |
+
`.pth` files are PyTorch pickles. Verify the checksums after download and load them only from this repository:
|
| 36 |
+
|
| 37 |
+
```bash
|
| 38 |
+
sha256sum pixelseal.pth pixelseal.yaml videoseal_attenuation.yaml audioseal_generator.pth audioseal_detector.pth
|
| 39 |
+
```
|
| 40 |
+
|
| 41 |
+
## Usage
|
| 42 |
+
|
| 43 |
+
### Install
|
| 44 |
+
|
| 45 |
+
Clone [LTX-2](https://github.com/Lightricks/LTX-2) and install the marking extras:
|
| 46 |
+
|
| 47 |
+
```bash
|
| 48 |
+
uv sync --extra videoseal --extra audioseal
|
| 49 |
+
# or
|
| 50 |
+
pip install 'ltx-pipelines[videoseal,audioseal]'
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
### Download the pack
|
| 54 |
+
|
| 55 |
+
The LTX-2 library never downloads weights itself. Fetch this repository to a local directory:
|
| 56 |
+
|
| 57 |
+
```bash
|
| 58 |
+
hf download Lightricks/LTX-2-Marking --local-dir path/to/marking
|
| 59 |
+
```
|
| 60 |
+
|
| 61 |
+
### Mark during generation
|
| 62 |
+
|
| 63 |
+
Video-generation and editing CLIs, HDR IC-LoRA, and text-to-audio accept `--marker-root-path`. Omit the flag to leave pixels and PCM unchanged.
|
| 64 |
+
|
| 65 |
+
```bash
|
| 66 |
+
python -m ltx_pipelines.distilled \
|
| 67 |
+
... \
|
| 68 |
+
--marker-root-path path/to/marking \
|
| 69 |
+
--output-path out.mp4
|
| 70 |
+
```
|
| 71 |
+
|
| 72 |
+
Pipeline-specific behavior:
|
| 73 |
+
|
| 74 |
+
- **HDR:** video marking is skipped; audio is still marked.
|
| 75 |
+
- **Audio-to-video:** only the video is marked, because the source audio is passed through.
|
| 76 |
+
- **Text-to-audio:** only the audio marker is used.
|
| 77 |
+
|
| 78 |
+
### Detect
|
| 79 |
+
|
| 80 |
+
```python
|
| 81 |
+
import torch
|
| 82 |
+
from ltx_pipelines.marking import create_markers, detect_audio_mark, detect_video_mark
|
| 83 |
+
|
| 84 |
+
video_marker, audio_marker = create_markers("path/to/marking", torch.device("cuda"))
|
| 85 |
+
print(detect_video_mark(video_marker, "out.mp4"))
|
| 86 |
+
print(detect_audio_mark(audio_marker, "out.mp4"))
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
Each call returns `MarkDetectResult(detected, bit_accuracy, p_value, payload)`. `bit_accuracy` and `p_value` come from a binomial test of the decoded bits against the payload, which is fixed in the LTX-2 code (not in these weights). `detected` is each marker's own decision:
|
| 90 |
+
|
| 91 |
+
- **Video:** every 8th decoded frame is resized to 256×256 and run through the extractor; bit logits are averaged over the sampled frames. `detected` is `log10(p_value) < -6`.
|
| 92 |
+
- **Audio:** the full track is scanned, with each stereo channel processed as its own mono signal. `detected` means at least half of the samples have a watermark probability above 0.5.
|
| 93 |
+
|
| 94 |
+
Detection runs on CPU or GPU. A GPU is much faster: on one B200, video detect takes about 15–30 ms per 5–10 s clip at up to 4K, versus 0.6–3 s on CPU. Both figures exclude file decoding.
|
| 95 |
+
|
| 96 |
+
## Scope and limitations
|
| 97 |
+
|
| 98 |
+
- Video marking targets SDR H.264 output; HDR output is not video-marked.
|
| 99 |
+
- AudioSeal runs at the native 24/48 kHz rate and does not resample.
|
| 100 |
+
- These are Meta's general-purpose models. Anyone holding the weights and the LTX-2 code can run detection, and a determined adversary may be able to weaken or remove the mark.
|
| 101 |
+
|
| 102 |
+
## Provenance
|
| 103 |
+
|
| 104 |
+
- `pixelseal.pth` is Meta's public PixelSeal checkpoint ([facebookresearch/videoseal](https://github.com/facebookresearch/videoseal)), repacked to the `model` state dict only. The optimizer, discriminator, EMA, and scheduler state were dropped; the model tensors are unchanged.
|
| 105 |
+
- `pixelseal.yaml` is the PixelSeal card from `videoseal` commit [`870ca7f`](https://github.com/facebookresearch/videoseal/tree/870ca7fb33578b90f14c602016b6c2788096226e). `videoseal_attenuation.yaml` is the `jnd_1_1` entry of that repo's `configs/attenuation.yaml`, which the pip package does not ship.
|
| 106 |
+
- `audioseal_generator.pth` and `audioseal_detector.pth` are the unmodified 16-bit base models from [facebook/audioseal](https://huggingface.co/facebook/audioseal).
|
| 107 |
+
|
| 108 |
+
## License
|
| 109 |
+
|
| 110 |
+
All weights and configs are released by Meta Platforms, Inc. under the MIT License; see `LICENSE-videoseal` and `LICENSE-audioseal`. The LTX-2 code is licensed separately; see the [LTX-2 repository](https://github.com/Lightricks/LTX-2).
|
| 111 |
+
|
| 112 |
+
## Citation
|
| 113 |
+
|
| 114 |
+
```bibtex
|
| 115 |
+
@article{fernandez2024video,
|
| 116 |
+
title = {Video Seal: Open and Efficient Video Watermarking},
|
| 117 |
+
author = {Fernandez, Pierre and Elsahar, Hady and Yalniz, I. Zeki and Mourachko, Alexandre},
|
| 118 |
+
journal = {arXiv preprint arXiv:2412.09492},
|
| 119 |
+
year = {2024}
|
| 120 |
+
}
|
| 121 |
+
|
| 122 |
+
@inproceedings{sanroman2024proactive,
|
| 123 |
+
title = {Proactive Detection of Voice Cloning with Localized Watermarking},
|
| 124 |
+
author = {San Roman, Robin and Fernandez, Pierre and Elsahar, Hady and D{\'e}fossez, Alexandre and Furon, Teddy and Tran, Tuan},
|
| 125 |
+
booktitle = {International Conference on Machine Learning (ICML)},
|
| 126 |
+
year = {2024}
|
| 127 |
+
}
|
| 128 |
+
```
|
audioseal_detector.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8a78e8a83584113523e161fc599fcab10fd0e94c04d2eb9d2fa1e9ec91ab69d9
|
| 3 |
+
size 34667641
|
audioseal_generator.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7a845b5fbe9364a63a3909d8ab3fe064d13a76ae4c2e983573e08c69b7b51748
|
| 3 |
+
size 58805980
|
pixelseal.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:97cff15f75515259b49a2cffc8fcd600c287cafe3f29c1d5c6ef9c81b492c088
|
| 3 |
+
size 308951307
|
pixelseal.yaml
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
checkpoint_path: https://dl.fbaipublicfiles.com/videoseal/pixelseal/checkpoint.pth
|
| 2 |
+
args:
|
| 3 |
+
attenuation_config: configs/attenuation.yaml
|
| 4 |
+
attenuation: jnd_1_1
|
| 5 |
+
nbits: 256
|
| 6 |
+
hidden_size_multiplier: 1.0
|
| 7 |
+
img_size_proc: 256
|
| 8 |
+
blending_method: additive
|
| 9 |
+
scaling_w: 0.2
|
| 10 |
+
scaling_i: 1.0
|
| 11 |
+
videowam_chunk_size: 32
|
| 12 |
+
videowam_step_size: 8
|
| 13 |
+
embedder:
|
| 14 |
+
model: unet_base_yuv_quant
|
| 15 |
+
params:
|
| 16 |
+
msg_processor:
|
| 17 |
+
nbits: 16
|
| 18 |
+
hidden_size: 32
|
| 19 |
+
msg_processor_type: 'binary+concat'
|
| 20 |
+
unet:
|
| 21 |
+
in_channels: 1
|
| 22 |
+
out_channels: 1
|
| 23 |
+
z_channels: 16
|
| 24 |
+
num_blocks: 8
|
| 25 |
+
activation: 'relu'
|
| 26 |
+
normalization: 'batch'
|
| 27 |
+
z_channels_mults: [2, 4, 8, 16]
|
| 28 |
+
last_tanh: True
|
| 29 |
+
extractor:
|
| 30 |
+
model: convnext_tiny
|
| 31 |
+
params:
|
| 32 |
+
encoder:
|
| 33 |
+
depths: [3, 3, 9, 3]
|
| 34 |
+
dims: [96, 192, 384, 768]
|
| 35 |
+
pixel_decoder:
|
| 36 |
+
pixelwise: False
|
| 37 |
+
upscale_stages: [1]
|
| 38 |
+
embed_dim: 768
|
| 39 |
+
nbits: 16
|
| 40 |
+
sigmoid_output: False
|
videoseal_attenuation.yaml
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# JND channel map from Meta videoseal configs/attenuation.yaml (not shipped in the pip package).
|
| 2 |
+
jnd_1_1:
|
| 3 |
+
in_channels: 1
|
| 4 |
+
out_channels: 1
|