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
Download pixelseal.yaml from Lightricks/LTX-2-Marking: direct link, hf CLI and curl.
- Browser
- Download file 940 Bytes
-
https://huggingface.co/Lightricks/LTX-2-Marking/resolve/main/pixelseal.yaml
- Command line
-
hf download hf://Lightricks/LTX-2-Marking/pixelseal.yaml
-
curl -L -o pixelseal.yaml https://huggingface.co/Lightricks/LTX-2-Marking/resolve/main/pixelseal.yaml
940 Bytes
| checkpoint_path: https://dl.fbaipublicfiles.com/videoseal/pixelseal/checkpoint.pth | |
| args: | |
| attenuation_config: configs/attenuation.yaml | |
| attenuation: jnd_1_1 | |
| nbits: 256 | |
| hidden_size_multiplier: 1.0 | |
| img_size_proc: 256 | |
| blending_method: additive | |
| scaling_w: 0.2 | |
| scaling_i: 1.0 | |
| videowam_chunk_size: 32 | |
| videowam_step_size: 8 | |
| embedder: | |
| model: unet_base_yuv_quant | |
| params: | |
| msg_processor: | |
| nbits: 16 | |
| hidden_size: 32 | |
| msg_processor_type: 'binary+concat' | |
| unet: | |
| in_channels: 1 | |
| out_channels: 1 | |
| z_channels: 16 | |
| num_blocks: 8 | |
| activation: 'relu' | |
| normalization: 'batch' | |
| z_channels_mults: [2, 4, 8, 16] | |
| last_tanh: True | |
| extractor: | |
| model: convnext_tiny | |
| params: | |
| encoder: | |
| depths: [3, 3, 9, 3] | |
| dims: [96, 192, 384, 768] | |
| pixel_decoder: | |
| pixelwise: False | |
| upscale_stages: [1] | |
| embed_dim: 768 | |
| nbits: 16 | |
| sigmoid_output: False | |