Instructions to use Vayden/LTX-2.5-MLX-Q8-Paged with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use Vayden/LTX-2.5-MLX-Q8-Paged with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir LTX-2.5-MLX-Q8-Paged Vayden/LTX-2.5-MLX-Q8-Paged
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
LTX-2.5 MLX Q8 paged components
Prepared components for WeeTodd Nodes and WeeTodd Studio on Apple Silicon. Download this package to reuse the prepared transformer and Gemma pages without local quantization. Keep all page manifests, notices and support files together.
Layout
transformer/: distilled transformer,weetodd-ltx25-transformer-paged-q8-v1.gemma/: Gemma 4 12B with LTX projections,weetodd-ltx25-gemma-paged-q8-v1.vae/ltx-2.5-video-vae-conv-bf16.safetensors: unchanged official convolutional video VAE.vae/ltx-2.5-audio-vae-bf16.safetensors: unchanged official audio VAE and vocoder.latent_upscale_models/ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0.safetensors: unchanged official spatial upscaler.
In Studio, choose an LTX 2.5 setup preset, select this downloaded directory and scan for the components. Existing components can also be selected independently. The compatible baseline uses distilled two-stage generation, staged unloading, Q8 paging/low-RAM streaming, and the reference-FP32 feed-forward path. This package contains neither IC-LoRAs nor guided/DFR development transformers.
Q8 is an approximation of BF16. File integrity was checked before publication; this release does not establish quality or memory suitability for every Mac, resolution, duration or media input. No new render was run during publication.
Sources and licenses
Immediate source files are from Lightricks/LTX-2.5, revision 5e6e71018ee1756ed329b697a7b4aedc934dfce9. The original transformer and encoder filenames are retained in the paging manifests. Unmodified VAE/upscaler hashes match that official revision. See MODIFICATIONS.md for all conversions and NOTICE for attribution.
The complete LTX-2.x Community License is included as LICENSE, including its attachment. All use and distribution restrictions apply, including Section 4 and Attachment A. The linked acceptable-use-policy snapshot is included as LTX_ACCEPTABLE_USE_POLICY.pdf. Commercial Entities must obtain the separate paid license required by the agreement before covered use; repository access provides no additional rights. Review the full terms and their exceptions.
Google Gemma 4 is Apache-2.0, recorded in GEMMA_APACHE_LICENSE and Google's official license page. The immediate Gemma pack includes Lightricks projections and remains distributed with the LTX source terms. No upstream rights are waived or replaced.
WeeTodd is independent of Lightricks and Google.
Quantized
Model tree for Vayden/LTX-2.5-MLX-Q8-Paged
Base model
Lightricks/LTX-2.5