Skip to content

README.md

Latest commit

 

History

History
47 lines (31 loc) · 3.05 KB

README.md

File metadata and controls

47 lines (31 loc) · 3.05 KB
Raw

LTX Audio and Video Generation

English | 简体中文

Prepare

Complete the environment setup, then choose an entry for the model version and task:

Version or task Entry
LTX-2 text to audio-video run_ltx2_t2av.sh
LTX-2 image to audio-video run_ltx2_i2av.sh
LTX-2.3 text / image to audio-video run_ltx2_3_t2av.sh / run_ltx2_3_i2av.sh
LTX-2.3 audio, keyframes, and motion conditioning ltx2_3
LTX-2.5 distilled, eight GPUs run_ltx2_5_t2av_distilled.sh / run_ltx2_5_i2av_distilled.sh

Download the corresponding LTX-2, LTX-2.3, or LTX-2.5 weights. In the selected Shell script, set lightx2v_path, model_path, and CUDA_VISIBLE_DEVICES, then edit the prompt and input paths.

  • LTX-2 / LTX-2.3: the local model_path directory must also contain the Gemma text encoder weights, tokenizer, and processor files. Download Gemma 3 into a subdirectory if needed. Set dit_original_ckpt in the selected JSON to the matching checkpoint. The LTX-2.3 scripts reuse the LTX-2 directory for Gemma and select the LTX-2.3 checkpoint through JSON.
  • LTX-2.5: set the component checkpoint paths in the selected JSON, including the Transformer, text encoder, video/audio VAE, duration head, and upsampler. The launchers use ltx2_5_distilled_8gpu.json.

Run

For the basic LTX-2 example, update dit_original_ckpt in ltx2.json, then run from the repository root:

bash scripts/ltx2/run_ltx2_t2av.sh

The basic text/image scripts save results to save_results/output_lightx2v_ltx2_t2av.mp4 or save_results/output_lightx2v_ltx2_i2av.mp4. Change --save_result_path to choose another location. Set resolution, frame count, and sampling parameters in the selected JSON under configs/ltx2.

For multi-GPU LTX-2 inference, choose the launchers ending in _cfg_parallel.sh, _ulysses.sh, or _tp.sh, and keep GPU count consistent with --nproc_per_node and the JSON parallel settings.

Python API

LTX-2 weights Text to audio-video Image to audio-video
Base ltx2_t2av.py ltx2_i2av.py
Distilled FP8 ltx2_t2av_distilled_fp8.py ltx2_i2av_distilled_fp8.py

Edit the model directory, checkpoint, input, and output paths in the selected example, then run:

python scripts/ltx2/inference/ltx2_t2av.py

See the Python API guide for shared usage.