LTX 2 now features native audio support, including Ukrainian, and has been released as open source. The audio quality is, to put it mildly, not optimal, and the model’s overall quality is among the lower end in recent months. However, this open-source project can be run at home, and it is the second neural network capable of generating Ukrainian speech. The release provides full access: weights, training code, and pipelines. It works on consumer graphics cards, with ComfyUI supported from the first day.
The model can generate videos with audio (dialogues, music, SFX) up to 20 seconds long. It produces output that appears as if in native 4K resolution, though it still involves upscaling rather than pure generation. Ukrainian language support is included; it is only the second neural network after Veo attempting to speak in Ukrainian, with some errors. Veo remains better, but this model’s attempt is noteworthy.
Where to try it (from Ukraine via VPN):
Where to download:
https://app.ltx.studio/ltx-2-playground/i2v
https://replicate.com/lightricks/ltx-2-distilled
https://fal.ai/models/fal-ai/ltx-2-19b/extend-video/lora
https://huggingface.co/Lightricks/LTX-2
https://github.com/Lightricks/LTX-2
https://www.youtube.com/@shtuchka-intelekt