SceneWorks/ltx-2.3-mlx

LTX-2.3 — MLX (self-contained: video model + Gemma-3-12B text encoder)

Native MLX (Apple Silicon) bundle for the SceneWorks app. Contains everything the LTX-2.3 MLX engine loads at generation time, in one repo:

dir	contents
q4/	LTX-2.3 distilled video model, MLX Q4 — transformer + connector + upsampler + vae_decoder + vae_encoder + audio_vae + vocoder + configs (split_model.json / embedded_config.json / config.json)
q8/	same components, MLX Q8 (higher quality, larger)
gemma/	Gemma-3-12B-IT text encoder, MLX bf16 (5 shards + tokenizer + config)

The engine reads the quant bits from each variant's split_model.json, and resolves the text encoder from gemma/.

Provenance

• LTX-2.3 video weights (q4/, q8/): converted from Lightricks/LTX-2.3 (distilled checkpoint) to MLX (Q4 / Q8) via the SceneWorks native mlx-gen-ltx converter.
• Gemma-3-12B text encoder (gemma/): mlx-community/gemma-3-12b-it-bf16 (MLX conversion of google/gemma-3-12b-it).

Licenses — this bundle contains TWO models under TWO licenses

• LTX-2.3 (q4/, q8/): LTX-2 Community License Agreement — https://github.com/Lightricks/LTX-2/blob/main/LICENSE . These are derivatives (MLX conversion / quantization) of LTX-2.3; redistribution is permitted under that license, which must travel with the weights, with the modification noted and the Attachment A use-restrictions applying. Entities with ≥ $10,000,000 annual revenue require a separate Lightricks commercial license.
• Gemma (gemma/): Google Gemma Terms of Use — https://ai.google.dev/gemma/terms . The Gemma Terms and the Gemma Prohibited Use Policy apply and must travel with these weights.

By downloading or using this repository you agree to both licenses.