LTX-2.5 (local Diffusers)

Open-weight LTX-2.5 via Lightricks/LTX-2.5-Diffusers. Runs on your CUDA GPU with the distilled recipe (explicit sigma schedule, CPU offload by default). Outputs MP4 with synchronized audio.

Hosted API twin: LTX-2.5 (official API) (--model ltx-2.5-api).

Requirements

Item	Notes
GPU	CUDA required; ~16GB VRAM minimum with offload; 24GB+ preferred
Disk	~72GB for distilled snapshot (exclude transformer_full/ if downloading manually)
License	LTX-2.x Community License — free commercial use under $10M ARR; accept HF gated terms
Diffusers	Must install from main — LTX-2.5 is not in a stable Diffusers release yet

pip install opentryon[local]
pip install "git+https://github.com/huggingface/diffusers"
export HF_TOKEN=hf_...   # after accepting the model card terms

Optional overrides:

# export LTX_MODEL_ID=Lightricks/LTX-2.5-Diffusers
# export LTX_MODEL_PATH=/path/to/local/snapshot

CLI

opentryon video-generate --model ltx-2.5 \
  --prompt "A fashion model walking a runway at dusk, camera tracking, soft ambient sound" \
  --width 960 --height 544 --num-frames 121 --frame-rate 24 --seed 42

opentryon video-generate --model ltx-2.5 \
  --image look.jpg \
  --prompt "Gentle fabric motion as the model turns, atelier ambience" \
  --width 960 --height 544 --num-frames 121

width/height must be divisible by 32. num_frames must satisfy num_frames % 8 == 1 (e.g. 97, 121).

Python

from tryon.models import LTX25Adapter

adapter = LTX25Adapter()  # loads Lightricks/LTX-2.5-Diffusers
video = adapter.generate_text_to_video(
    prompt="A cinematic shot of a red fox walking through snow, camera tracking",
    width=960,
    height=544,
    num_frames=121,
    seed=42,
)
open("ltx25.mp4", "wb").write(video)

Notes

• Default path uses the distilled transformer + convolutional VAE decode (same as the HF quick start).
• Prompts work best as long single-paragraph audiovisual captions.
• For production Playground metering, wrap this adapter in a GPU job worker — do not load weights inside Django.