Hunting LTX-2 weights? Here’s the field guide.

What it does This repository is a curated index of models, encoders, and tools for Lightricks’ LTX-2 open-source video generation suite. It catalogs official and community checkpoints across multiple precisions—bf16, fp8, int8, NVFP4, and fp4—plus distilled LoRAs, spatial upscalers, and third-party desktop wrappers. Think of it as a shopping list for ComfyUI users trying to avoid downloading the wrong 46 GB file.

The interesting bit The list goes beyond simple links: it tracks experimental LoRA variants with ranks as low as 36, specifically tuned for image-to-video workflows, and flags which quantizations require custom ComfyUI forks. That level of granularity is usually buried in Discord threads, not a README.

Key highlights

• Curated index of LTX-2 (19B) and LTX-2.3 (22B) checkpoints with direct download links.
• Tracks aggressive quantization options, from ~17 GB NVFP4 distilled weights to fp8_e5m2 builds for older 30-series GPUs.
• Catalogs distilled LoRAs from massive rank-384 official files down to compact 36-rank community experiments.
• Flags compatibility gotchas, such as mxfp8mixed quantizations requiring a custom ComfyUI-Kitchen fork.
• References third-party tools like LTX2.3-Multifunctional, a desktop app claiming to cut VRAM requirements from 32 GB to 24 GB.

Caveats

• The repository is purely a reference index; it contains no code, nodes, or automation.
• Several quantization formats require non-standard ComfyUI forks or specific hardware generations.
• The README is truncated mid-table, so sections on encoders and workflows are incomplete.

Verdict Worth bookmarking if you are actively experimenting with LTX-2 in ComfyUI and tired of hunting Hugging Face links. If you are not already committed to the LTX-2 ecosystem, this list will not convince you to join.