Model weights for the LTX 2.5 inference path (engine25) of
Nz-Videomni (formerly Nz-LTX23), a plugin
for the video editor AviUtl2.
Three of the four files (both VAEs, the upscaler) are verbatim byte-for-byte re-hosts of
upstream LTX 2.5 files. The transformer file is a GGUF quantization produced locally by
Rootport from the upstream bf16 weights, using
Nz-GGUF-Converter-LTX23 (tool 1.2.0,
commit 905f348): 1,658 of 4,349 tensors are Q4_K-quantized; the remaining 2,691 tensors
(2,401 BF16 + 290 F32) are passed through unmodified. See NOTICE.md for the full
provenance table.
The text encoder (a GGUF quantization of the LTX-fine-tuned Gemma 4) is hosted separately at
Rootport/Nz-Gemma4-12B-LTX25 because its
base model carries a different license (Apache 2.0).
Not compatible with LTX 2.3 weights (Rootport/Nz-LTX23-weights) — the two versions live in
separate directories (models/LTX23/ and models/LTX25/) but their files cannot be mixed.
License: LTX-2.x Community License Agreement, dated August 11, 2026 (LICENSE);
use restrictions (Section 4 / Attachment A), the transfer-notice obligations of Section 3.5, and the
AI-regulation obligations of Section 6 are reproduced in NOTICE.md. Entities with
annual revenues of at least $10,000,000 must obtain a separate commercial license from Lightricks
Ltd. Not an official Lightricks distribution, and not endorsed by Lightricks Ltd.