ltx-2-3-demo

LTX-2.3 Demo

A Gradio Space demonstrating Lightricks/LTX-2.3, a 22B-parameter DiT-based joint audio-video foundation model. Supports:

• Text-to-video — just enter a prompt.
• Image-to-video — also upload a conditioning image to animate it.

Both produce short video clips with synchronized audio.

Hardware

This is a large (22B parameter) model. It needs a GPU with substantial VRAM (A100 80GB / H100 class). It will not run on CPU or on small/free GPU tiers. If you're deploying this Space yourself, set the Space hardware accordingly in Settings → Hardware.

If that's too expensive/slow for your use case, swap MODEL_ID in app.py for one of the lighter distilled checkpoints listed on the model card (e.g. ltx-2.3-22b-distilled, 8 steps, CFG=1).

Files

• app.py — Gradio UI + inference logic (diffusers DiffusionPipeline).
• requirements.txt — installs diffusers from source, since native LTX-2.3 support is very recent.

Notes

• Diffusers support for LTX-2.3 is brand new; if from_pretrained fails, check the model card for the latest recommended pipeline call, or fall back to the official LTX-2 GitHub pipelines instead.
• Frame counts must be divisible by 8, plus 1 (e.g. 97, 121, 161) per the model's requirements. Height/width must be divisible by 32.