Downloads · 30 days
0
o2noor/comfyui-LTX-2.5-Tile-train-LoRa-On-multi-Gpus-low-VRAM-18-gb-Beta
comfyui-LTX-2.5-Tile-train-LoRa-On-multi-Gpus-low-VRAM-18-gb-Beta is a text-to-video model from o2noor. Use it when you need video from a text prompt. The card lists the license as other.
Quantized + auxiliary weights for the O2noor LTX-2.5 Int4 Tile-Train ComfyUI training pack (22B LTX-2.5 distilled, bnb-NF4). Download these into your ComfyUI model folders.
Downloads · 30 days
0
Access
Public
Updated Aug 24, 2026
Repo size
55.8 GB
Likes
4
Trending 1
Click a slice to open those files.
.safetensors55.8 GB · 100%
From the Hugging Face model README
Quantized + auxiliary weights for the O2noor LTX-2.5 Int4 Tile-Train ComfyUI training pack (22B LTX-2.5 distilled, bnb-NF4). Download these into your ComfyUI model folders.
| File | Size | Put in |
|---|---|---|
| ltx-2.5-22b-distilled-bnb-nf4.safetensors | 10.4 GB | diffusion_models/ |
| embeddings_processor_bf16.safetensors | 6.3 GB | diffusion_models/ |
| gemma4-12b-with-proj-ltx-2.5-bf16.safetensors | 26 GB | text_encoders/ |
| ltx-2.5-video-vae-bf16.safetensors | 1.5 GB | vae/ |
| ltx-2.5-audio-vae-bf16.safetensors | 0.36 GB | vae/ |
The base model is fully self-contained - no separate connectors file or quantized cache folders needed.
The ~26 GB text encoder runs 8-bit (LLM.int8) spread across 2 GPUs during captioning/generation and is freed; it is not loaded during training (training uses cached embeddings). It works fine on 12 GB cards.
ComfyUI node pack (clone, run install.py, place the 5 models, load the workflow):
https://github.com/A4ax/comfyui-LTX-2.5-Tile-train-LoRa--On-multi-Gpus-low-VRAM-18-gb-Beta
These weights are derivatives of LTX-2.5 and remain subject to the LTX-2.x Community License + Attachment A. See the GitHub repo README -> "Licensing / Model redistribution" for full terms.